首页|Deep learning-based integrated grid and section optimization for Kiewit shell structures
Deep learning-based integrated grid and section optimization for Kiewit shell structures
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
Elsevier
The optimal design of Kiewit shell structures typically involves two key aspects: grid optimization and section optimization. However, traditional design software primarily designs member sections through finite element analysis and iterative optimization under predetermined grid conditions, making it difficult to achieve a global coupling between the grid form and section configuration. To address this limitation, this paper proposes an integrated design method assisted by deep learning. By setting an overall optimization objective, structural features are automatically extracted from a large database of design schemes, and the optimal grid along with its corresponding member sections is simultaneously selected. Initially, the paper introduces the core concepts of the proposed method, including the multi-feature embedding unit, the embedding encoding principle, and the overall framework of the single-layer spherical lattice shell design model. Subsequently, using an optimal grid selection technique, candidate structures with minimal steel consumption are identified under various design conditions, with spans ranging from 30 to 80 m. To validate the effectiveness of the proposed approach, a comparative analysis with traditional PKPM software is conducted. Experimental results indicate that the deep learning-assisted integrated optimization method not only generates grid forms and section configurations that meet code requirements but also offers significant advantages in reducing steel consumption, enhancing material utilization, and improving design efficiency.
NETL
NSTL
Elsevier
