摘要
传统十字板式节点应用广泛,但样式单一且材料布局不尽合理,为此,提出了一种基于边界平衡生成对抗网络(BEGAN)的十字板式节点新构形智能生成方法,可以生成多种造型新颖且性能更优的节点构形。基于变密度法定义最大化刚度为优化目标,通过调整优化参数获取大量不同工况下的十字板式节点拓扑优化结果,建立深度学习数据集。基于开源深度学习框架TensorFlow搭建BEGAN,深度学习数据集样本的真实分布,通过定义标签类型智能生成相应工况下的节点新方案。评估分析生成方案的新颖性、多样性、力学性能与质量,并借助三维重构技术与增材制造技术分析了智能生成节点的制造可行性与美观性。研究结果表明:该方法经过25次迭代训练即可稳定收敛,不仅能够高效地批量化智能生成创新性突出的若干节点设计新方案,满足结构设计的多元审美要求,而且能够优化节点的力学性能与材料用量,解决十字板式节点的材料布局不尽合理问题。
The traditional cross-plate joint is widely applied in practical engineering,but it has a single style and unreasonable material layout.Therefore,an intelligent generation method for innovative configurations of the cross-plate joint based on boundary equilibrium generation adversarial networks(BEGAN)is proposed,which can generate a variety of innovative joints with novel configurations and better performance.The maximum stiffness is defined as the optimization objective based on the variable density method.By adjusting the optimization parameters,a large number of topology optimization results of the cross-plate joint under different load cases are obtained for building a deep learning dataset.Based on the open-source deep learning framework TensorFlow,the BEGAN algorithm is programmed to deeply learn the real distribution of the dataset samples,it can intelligently generate design schemes under the corresponding working conditions by defining the label categories.The novelty,diversity,mechanical properties and mass of the generated schemes are evaluated,and the manufacturing feasibility and aesthetics of the intelligently generated joint are analyzed with the help of 3 D reconstruction and additive manufacturing technology.The results show that the approach converges stably after 25 epochs of iterative training,and can not only efficiently and intelligently generate innovative joint solutions in batch,which meets the multiple aesthetic requirements of structural design,but also optimize the mechanical properties and material usage of the joint,and solve the problem of the unreasonable material layout of the cross-plate joint.
作者
杜文风
王英奇
王辉
赵艳男
高博青
董石麟
DU Wenfeng;WANG Yingqi;WANG Hui;ZHAO Yannan;GAO Boqing;DONG Shilin(Institute of Steel and Space Structures,Henan University,Kaifeng 475004,China;Department of Civil Engineering,Zhejiang University,Hangzhou 310058,China)
出处
《建筑结构学报》
EI
CAS
CSCD
北大核心
2022年第S01期315-324,共10页
Journal of Building Structures
基金
国家自然科学基金项目(U1704141,52178172)
河南省高校科技创新团队支持计划项目(22IRTSTHN019)
浙江省空间结构重点实验室开放基金资助项目(202106)
关键词
十字板式节点
边界平衡生成对抗网络
拓扑优化
构形
增材制造
cross-plate joint
boundary equilibrium generative adversarial networks
topology optimization
configuration
additive manufacturing