A review on artificial intelligence applications for facades

Abstract

This review applies a transformer-based topic model to reveal trends and relationships in Artificial Intelligence (AI)-driven facade research, with a focus on architectural, environmental, and structural aspects. AI methods reviewed include Machine Learning (ML), Deep Learning (DL), and Computer Vision (CV). Overall, a significantly growing interest in applying AI methods can be observed across all research areas. However, noticeable differences exist between the three topics. While CV and DL techniques are applied to image data in research on the architectural design of facades, research on environmental aspects of facades often uses numerical data with relatively small datasets and classical ML models. Research on facade structure also tends to use image data but also incorporates numerical performance prediction. A major limitation remains a lack of generalizability, which could be addressed by more comprehensive datasets and novel DL techniques. These include concepts such as Physics-Informed Neural Networks, where domain knowledge is integrated into hybrid data-driven models, and multi-modal diffusion models, which offer generative modeling capabilities to support inverse and forward design tasks. The trends and directions outlined in this review suggest that AI will continue to advance facade research and, in line with other domains, has the potential to achieve a level of maturity suitable for adoption beyond academia and into practice. • A review on AI applications for building facades using a transformer-based topic model. • Deep Learning and Computer Vision are popular in architectural and structural topics. • Classical Machine Learning is more common in environmental applications with limited data. • Major challenges exist in addressing generalizability and data set quality and size. • Future directions involve multimodal datasets, physics-informed networks and generative AI.

Topics & Concepts