Research on Multimodal Semantic Understanding and 3D CAD Generation for Mechanical Design: From Understanding and Knowledge Modeling to Generation and Verification

Abstract

Mechanical design is entering a new era of multimodal and AI-enabled intelligence. Yet in engineering practice, current generative methods still struggle with limited multimodal fusion, weak modelling of complex design knowledge, and insufficient engineering usability of generated results. This poster presents our recent research on multimodal semantic understanding and 3D CAD generation for mechanical parts. We build an integrated framework that links semantic understanding, knowledge modelling, CAD generation, and result validation. The work includes fine-grained multimodal feature fusion for mechanical parts, automated construction of multimodal knowledge graphs, and engineering-oriented 3D/CAD generation methods based on diffusion models and constraint-aware design strategies. Our goal is to enable AI not only to “understand” mechanical parts, but also to generate editable, manufacturable, and application-ready CAD models. This research offers a practical path toward more efficient, reliable, and intelligent design for complex mechanical products.

Speaker

Prof. Wang Zuoxu is an Associate Professor at the School of Mechanical Engineering and Automation, Beihang University. She has been engaged in research and practice in intelligent design and manufacturing,  knowledge graphs for the manufacturing industry, and AIGC in engineering design. She has undertaken national and provincial-level research projects, including the National Natural Science Foundation Youth Program and sub-projects of key R&D initiatives, and participated in major national projects. In recent years, she has co-authored two academic monographs and published over 40 high-level academic papers and two authorized invention patents.