Deep learning for safety risk management in modular construction: Status, strengths, challenges, and future directions

Abstract

Occupational health risks such as falls from height, electrocution, object strikes, mechanical injuries, and collapses have plagued the construction industry. Deep learning algorithms are exploding due to their outstanding analytical capabilities and are believed to improve safety management significantly. Therefore, this paper systematically reviewed the literature on DL algorithms from 2015 to 2024 in modular construction. It found that the six most popular DL algorithms in this area are “Convolutional Neural Network (CNN),” “Recurrent Neural Network (RNN),” “Generative Adversarial Network (GAN),” “Auto-Encoder (AE),” “Deep Belief Network (DBN)” and “Transformer.” However, in addition to each algorithm's limitations, problems like data constraints, talent gaps, and a lack of guidance frameworks also exist. To address these issues, three strategies are proposed. They are “establishing a multi-modal data sharing platform,” “proposing a paradigm framework for the application of DL algorithms,” and “constructing a compound construction talent training mechanism,” which provide researchers with future references.