Intelligent design method for beam and slab of shear wall structure based on deep learning
Intelligent design method for beam and slab of shear wall structure based on deep learning
复制标题
基于深度学习的剪力墙结构梁板智能设计方法
DOI：
10.1016/j.jobe.2022.104838
复制
发表时间：
2022
影响因子：
6.4
通讯作者：
Xinzheng Lu
中科院分区：
文献类型：
--
作者：
Pengju Zhao;Wenjie Liao;Hongjing Xue;Xinzheng Lu
梁板设计是剪力墙结构设计的重要组成部分。目前，传统的手工设计是耗时的，并且定义优化设计的目标函数和设计变量是具有挑战性的。相比之下，深度学习方法可以学习高维图像特征并生成新的设计，为高效智能的结构设计提供新的解决方案。为此，基于深度神经网络，提出了一种融合建筑空间和构件属性的钢筋混凝土剪力墙结构梁智能布局设计方法。该方法学习了已有设计的隐含规律，实现了新布局方案的推理生成。随后，基于数理统计，提出了确定连梁和框架梁类型和尺寸的方法。典型算例分析表明，按该方法设计的梁板结构性能与工程师设计的梁板结构性能相当。按该方法设计的结构最大层间位移比与工程师设计的结构最大层间位移比相差不超过5 × 10−5。计算结果与工程师设计结果相比，典型楼板最大竖向位移、混凝土用量和钢筋用量分别相差0.8%、2.88%和6.20%。设计效率显著提高30倍以上。
Beam and slab design is a critical component of shear wall structure design. Currently, conventional manual design is time-consuming, and defining objective functions and design variables of an optimization design is challenging. In contrast, deep learning methods can learn high-dimensional image features and generate new designs, providing new solutions for efficient and intelligent structural design. Therefore, based on deep neural networks, this study proposes an intelligent layout design method for beams of reinforced concrete shear-wall structures using the input of fused building space and element attributes. This method learned the implicit laws of existing designs and realized the inferential generation of new layout schemes. Subsequently, based on mathematical statistics, methods to determine the type and size of coupling and frame beams are proposed. A typical case study shows that the structural performance of the beam and slab designed by this method was comparable to that of competent engineers. The maximum inter-story drift ratio of the result designed by the proposed method differs from that designed by engineers by no more than 5 × 10−5. The differences in the maximum vertical typical-floor-slab displacement, the concrete consumption, and the steel consumption between the design result of the proposed method and the engineer's design result are 0.8%, 2.88%, and 6.20%, respectively. Moreover, the design efficiency was significantly improved by more than 30 times.