GAMES Webinar 2026 – 412期(智能建模与全六面体网格生成)|童话 (Carnegie Mellon University), 余宇轩(东华大学)

GAMES Webinar 2026 – 412期(智能建模与全六面体网格生成)

报告嘉宾:童话 Carnegie Mellon University

报告时间:2026年08月20日 晚上20:00-20:30(北京时间)

报告题目:

Advances in grid-based all-hexahedral mesh generation: saving elements and guaranteeing positive Jacobian

报告摘要:

In the CAE domain, hexahedral meshing is widely regarded as the “Holy Grail,” due to its superior numerical accuracy and difficulty relative to tetrahedral meshing. It remains the core bottleneck for Isogeometric Analysis. In this talk, I will present two theoretical contributions of my Ph.D. research, which target long-standing open problems in grid-based hexahedral meshing. The first work aims to reduce the number of background hexahedral cells under a given adaptive octree, so that future algorithms can scale up to handle more complex input geometries within limited memory budgets. The second work introduces the Marching Cubes algorithm into the meshing pipeline. It ensures that, prior to the nonlinear optimization in the final trimming step, the interior–exterior interface of the hexahedral mesh already converges to the input geometry at a faster rate, while also establishing a theoretical lower bound on mesh quality. This is the first hexahedral meshing algorithm that provides both a second-order convergence rate in Hausdorff distance and a guaranteed positive minimum Jacobian determinant.

讲者简介:
Hua Tong is currently a 4th year Ph.D. Candidate in Mechanical Engineering at Carnegie Mellon University, advised by Prof. Yongjie Jessica Zhang. He received his Bachelor’s degree in Theoretical and Applied Mechanics from the School of the Gifted Young at the University of Science and Technology of China in 2022. His research interests include volumetric mesh generation and surface reconstruction. He publishes in journals such as ACM Transactions on Graphics, Computer-Aided Design, and Computer-Aided Geometric Design. He is the recipient of several honors, including the International Meshing Roundtable 2025 Best Student Paper Award and the 2026 CMU Liang Ji-Dian Fellowship.
讲者主页:gt2001.github.io

报告嘉宾:余宇轩 东华大学

报告时间:2026年08月20日 晚上20:30-21:00(北京时间)

报告题目:

让 AI 生成可计算的三维实体:面向 CAD/CAE 一体化的智能建模

报告摘要:

等几何分析(IGA)采用统一的样条表示连接几何设计与数值分析,为 CAD/CAE 一体化提供了一条技术路径。当前,如何由给定的边界曲面构造高质量的三维体样条参数化模型,仍是等几何分析面临的主要挑战之一。本报告围绕多立方体参数域、全六面体控制网格、三维体样条和等几何分析构成的技术链条,介绍人工智能驱动的三维实体建模方法。前期研究开发了 HexGen/Hex2Spline 和 HexDom 等网格与样条建模工具,其中 HexGen/Hex2Spline 已被 Ansys LS-DYNA 采用,相关成果也服务于 Honda、GEM 等工程场景。在此基础上,报告重点介绍 DL-Polycube 和 DDPM-Polycube。DL-Polycube 利用图神经网络预测预定义的多立方体结构,并以此指导输入几何的无监督表面分割。DDPM-Polycube 将几何到多立方体结构的变形建模为去噪扩散过程,通过反向扩散生成有效的多立方体结构,降低对大型预定义模板库的依赖。相关方法结合参数映射、八叉树细分、网格质量优化和截断分层三维体样条构造,形成从 CAD 边界曲面到全六面体控制网格、三维体样条和等几何分析的自动化流程。最后,报告结合心血管时空等几何血流分析和人工智能赋能的 4D 打印,讨论几何建模、数值仿真与数据驱动方法的交叉应用。

讲者简介:

余宇轩,东华大学人工智能研究院讲师,美国 Carnegie Mellon University 机械工程系博士。本科毕业于复旦大学,硕士毕业于 Rice University。主要研究方向包括网格生成、三维体样条参数化、等几何分析、CAD/CAE 一体化,以及深度学习和生成式人工智能在工程建模中的应用。主持国家自然科学基金青年项目、上海市白玉兰人才计划浦江项目和教育部基金项目 2 项,并作为主要成员参与美国制造业未来启动项目、美国海军航空系统司令部项目和 Honda 汽车项目。在人工智能、神经网络与机器学习、计算机辅助设计、计算力学和人机交互等相关领域发表学术论文,相关成果见于 IEEE TNNLS、JMPT、ACM UIST、Eng Comput、Comput Mech、CAD 等期刊和会议,并担任多个 SCI 期刊审稿人。其 HexGen/Hex2Spline 工具已被 Ansys LS-DYNA 采用,相关成果服务于 Honda、GEM 等企业应用场景,并已应用于增材制造、工程设计和生物医学等领域。

讲者主页:https://iai.dhu.edu.cn/2025/0708/c20257a363745/page.htm


主持人简介:

肖艳阳,南昌大学数学与计算机学院副教授,博士毕业于厦门大学。主要关注网格生成、图像处理等方面的研究,在国内外重要期刊上发表和录用论文20余篇。主持国家自然科学基金和江西省自然科学基金项目。曾获SPM2018最佳论文一等奖。

You may also like...