Automatic tower crane layout planning system for high-rise building construction using generative adversarial network

• Published in: Advanced engineering informatics 2023-10, Vol.58, p.102202, Article 102202
• Main Authors: Li, Rongyan, Chi, Hung-Lin, Peng, Zhenyu, Li, Xiao, Chan, Albert P.C.
• Format: Article
• Language: English
• Subjects: Automatic design; Computer vision; Crane location; Generative adversarial network; Image-to-image translation; Tower crane
• ISSN: 1474-0346; 1873-5320
• DOI: 10.1016/j.aei.2023.102202

•An image-based automatic tower crane layout planning approach is proposed.•A highly scalable crane layout prediction network, CraneGAN, is developed.•The results of CraneGAN in transportation time outperformed GA in 66.67% of cases.•Taking images as input also saves manual data extraction time for planning preparation.•CraneGAN prediction remains in polynomial time while maintaining the layout quality. With the spring up of high-rise building projects, tower crane layout planning (TCLP) is increasingly crucial to avoid construction costs, safety issues, and productivity deficiencies. Current optimization approaches require manual data extraction and become more complex as projects scale growing. To further alleviate the planning burden, an automatic TCLP system is proposed, using a generative adversarial network (GAN) called CraneGAN. It generates tower crane layouts from drawing inputs, eliminating the need for manual information extraction. CraneGAN is trained on a high-quality dataset and evaluated based on its computational time and crane transportation time. By adjusting hyperparameters and applying data augmentation, CraneGAN achieves robust and efficient results compared to genetic algorithms (GA) and the exact analytics method. After validating through a numerical analysis for construction project, this proposed approach overcomes complexity limitations and streamlines the manual data extraction process to better facilitate layout planning decision-making.