Abstract：[Objective] Deep learning methods have mastered sophisticated structural design skills via feature extraction and data-driven learning, thus significantly advancing intelligent building structural design. The performance of data-driven intelligent structural design typically depends on the quantity and distribution of training data. However, only a few studies have investigated the relationship between the learned data features and the intelligent design, thus limiting its performance. This study aims to study the influence of training data distribution and quantity on structural design and subsequently improve intelligent design quality.[Methods] Two typical datasets are generated by collecting the shear wall architectural-structural design drawing data from two distinct regions (Beijing and Shandong) with different design habits. Based on the Shandong dataset, the data distribution characteristics are studied, and the correlation between the design conditions, the structural planer, the vertical shape, and the equivalent ratio of the shear wall to the architectural plan area are investigated using linear regression analysis. Subsequently, a generative adversarial network-based intelligent design method for building structures is adopted to extract and learn high-dimensional data features. According to the quantity and the distribution characteristics of data, this study has proposed a data augmentation method and a two-stage training method, wherein all data is hybridized for training in the first stage, and the data grouped by different design conditions is used in the second stage, thereby improving the design quality of intelligent design models. In addition, the Beijing dataset, which is substantially different from the training data, is used to evaluate and validate the study results using various training methods and training data quantities. Finally, to validate the results and illustrate the performance of the intelligent structural design, typical cases with three different design conditions are used, namely, 7 degree seismic intensity and 27 m structural height, 7 degree seismic intensity and 54 m structural height, and 8 degree seismic intensity and 39 m structural height.[Results] The analysis results demonstrated the following. 1) The regression analysis based on low-dimensional data features could not guide the design generation properly for complicated structures, such as shear walls. In contrast, high-dimensional feature learning based on deep learning might effectively capture the potential design laws and optimize the design generation. 2) With the improvement of the quantity of training data and the training strategy, the quality of the intelligent design structure increased by an average above 20%; however, the quality of the intelligent design was compromised when the properties of the test and training data were considerably different (with a shear wall ratio difference of over 50%). 3) Moreover, the analysis results were evaluated using relevant case studies. Regarding the plane's design and the mechanical performance of the overall structure, the intelligent design and the engineer's design had a high degree of resemblance, and the maximum interstory drift ratio varied by up to 8% on average.[Conclusions] Consequently, by assessing the data distribution, design conditions, and quantitative properties, the generative adversarial network-based intelligent structural design may provide high-quality designs with suitable training datasets. Furthermore, this study provides a reference for future research on intelligent structural design based on deep learning and the influence of data features.
刘元鑫, 廖文杰, 林元庆, 解琳琳, 陆新征. 数据特征对剪力墙结构生成式智能设计的影响[J]. 清华大学学报（自然科学版）, 2023, 63(12): 2005-2018.
LIU Yuanxin, LIAO Wenjie, LIN Yuanqing, XIE Linlin, LU Xinzheng. Influence of data features on the generative adversarial network-based intelligent design for shear wall structures. Journal of Tsinghua University(Science and Technology), 2023, 63(12): 2005-2018.
[1] DAVIES J. Smart contract operating project networks:Using integrated functionality for improving workflow, transparency, and easier collaboration[D]. Denmark:Via University College, 2018. [2] 周绪红,刘界鹏,冯亮,等.建筑智能建造技术初探及其应用[M].北京:中国建筑工业出版社, 2021. ZHOU X H, LIU J P, FENG L, et al. A preliminary exploration and application of building intelligent construction technology[M]. Beijing:China Architecture&Building Press, 2021.(in Chinese) [3] MáLAGA-CHUQUITAYPE C. Machine learning in structural design:An opinionated review[J]. Frontiers in Built Environment, 2022, 8:815717. [4] SUN H, BURTON H V, HUANG H L. Machine learning applications for building structural design and performance assessment:State-of-the-art review[J]. Journal of Building Engineering, 2021, 33:101816. [5] PIZARRO P N, HITSCHFELD N, SIPIRAN I, et al. Automatic floor plan analysis and recognition[J]. Automation in Construction, 2022, 140:104348. [6] CHANG K H, CHENG C Y. Learning to simulate and design for structural engineering[C]//International Conference on Machine Learning. PMLR, 2020:1426-1436. [7] CHANG K H, CHENG C Y, LUO J L, et al. Building-GAN:Graph-conditioned architectural volumetric design generation[C]//2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada:IEEE, 2021:11956-11965. [8] NAUATA N, HOSSEINI S, CHANG K H, et al. House-GAN++:Generative adversarial layout refinement network towards intelligent computational agent for professional architects[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville, USA:IEEE, 2021:13632-13641. [9] HAYASHI K, OHSAKI M. Graph-based reinforcement learning for discrete cross-section optimization of planar steel frames[J]. Advanced Engineering Informatics, 2022, 51:101512. [10] ZHU S J, OHSAKI M, HAYASHI K, et al. Machine-specified ground structures for topology optimization of binary trusses using graph embedding policy network[J]. Advances in Engineering Software, 2021, 159:103032. [11] JEONG J H, JO H. Deep reinforcement learning for automated design of reinforced concrete structures[J]. Computer-Aided Civil and Infrastructure Engineering, 2021, 36(12):1508-1529. [12] LIAO W J, LU X Z, HUANG Y L, et al. Automated structural design of shear wall residential buildings using generative adversarial networks[J]. Automation in Construction, 2021, 132:103931. [13] LU X Z, LIAO W J, ZHANG Y, et al. Intelligent structural design of shear wall residence using physics-enhanced generative adversarial networks[J]. Earthquake Engineering&Structural Dynamics, 2022, 51(7):1657-1676. [14] LIAO W J, HUANG Y L, ZHENG Z, et al. Intelligent generative structural design method for shear wall building based on "fused-text-image-to-image" generative adversarial networks[J]. Expert Systems with Applications, 2022, 210:118530. [15] FEI Y F, LIAO W J, ZHANG S, et al. Integrated schematic design method for shear wall structures:A practical application of generative adversarial networks[J]. Buildings, 2022, 12(9):1295. [16] ZHAO P J, LIAO W J, XUE H J, et al. Intelligent design method for beam and slab of shear wall structure based on deep learning[J]. Journal of Building Engineering, 2022, 57:104838. [17] 廖文杰,陆新征,黄羽立,等.剪力墙结构智能化生成式设计方法:从数据驱动到物理增强[J/OL].(2022-07-21)[2022-10-10]. http://kns.cnki.net/kcms/detail/50.1218.TU.20220720.1117.004.html. LIAO W J, LU X Z, HUANG Y L, et al. Intelligent generative structural design methods for shear wall buildings:From data-driven to physics-enhanced[J/OL].(2022-07-21)[2022-10-10]. http://kns.cnki.net/kcms/detail/50.1218.TU.20220720.1117.004.html.(in Chinese) [18] FEI Y F, LIAO W J, HUANG Y L, et al. Knowledge-enhanced generative adversarial networks for schematic design of framed tube structures[J]. Automation in Construction, 2022, 144:104619. [19] 程国忠,周绪红,刘界鹏,等.基于深度强化学习的高层剪力墙结构智能设计方法[J].建筑结构学报, 2022, 43(9):84-91. CHENG G Z, ZHOU X H, LIU J P, et al. Intelligent design method of high-rise shear wall structures based on deep reinforcement learning[J]. Journal of Building Structures, 2022, 43(9):84-91.(in Chinese) [20] PIZARRO P N, MASSONE L M, ROJAS F R, et al. Use of convolutional networks in the conceptual structural design of shear wall buildings layout[J]. Engineering Structures, 2021, 239:112311. [21] PIZARRO P N, MASSONE L M. Structural design of reinforced concrete buildings based on deep neural networks[J]. Engineering Structures, 2021, 241:112377. [22] WANG T C, LIU M Y, ZHU J Y, et al. High-resolution image synthesis and semantic manipulation with conditional GANs[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA:IEEE, 2018:8798-8807. [23] 龙湖地产集团总部.龙湖地产结构设计限额控制指标[R]. 2014. LONGFOR PROPERTIES CO LTD. Structure design quota control index of Longfor Properties Co Ltd.[R]. 2014.(in Chinese) [24] TAFRAOUT S, BOURAHLA N, BOURAHLA Y, et al. Automatic structural design of RC wall-slab buildings using a genetic algorithm with application in BIM environment[J]. Automation in Construction, 2019, 106:102901.
2023, Vol. 63 
