AI-assisted generative workflow for the early-stage design of sustainable tall buildings based on their structural behaviour

The construction industry is actively working toward reducing the carbon footprint related to the construction of tall buildings. For this purpose, the structural frame, a major cost factor, needs to efficiently merge with architectural aesthetics. This study utilises parametric design and machine learning (ML) algorithms to identify efficient structural solutions for tall buildings with outer diagrids. According to a supervised ML approach, data relevant to seismic simulations for a small dataset of building skin geometries are handled. ML tools are then used to explore the relationships between building features and responses, informing architectural and structural decisions. Alternative feature selection methods are employed to prioritize influential aspects. To enhance the generalisation capability of the procedure, outcomes from original numerical models are supplemented by data generated through a synthesis algorithm, the Gaussian copula (GC), which is used for data augmentation. The effectiveness of the random forest (RF) regressor is then demonstrated, even when applied to synthetic data, while maintaining a good performance thanks to cross-validation (CV) techniques. Although with a slight reduction in the correlation, as compared to the original dataset, the proposed data augmentation technique looks acceptable as a trade-off between efficiency and complexity reduction through data synthesis. Results demonstrate the machine learning capability to handle the augmented datasets, offering early-stage design insights for the desired outcomes.