Scaffold Safety Analysis: Focusing on Deep Learning

Sayan Sakhakarmi, University of Nevada, Las Vegas
Cristian Arteaga, University of Nevada, Las Vegas
Chunhee Cho, University of Hawaii at Manoa
Jee Woong Park, University of Nevada, Las Vegas


Recent advancements in data science have attracted construction researchers towards sophisticated techniques in data analytics, such as machine learning (ML), for the active control of construction sites. As such, researchers have applied ML to address various construction safety issues. Despite significant progress in this field, the applicability of deep learning in construction has not been studied sufficiently. Therefore, this paper focuses on exploring the capability of deep learning to address safety issues through learning data structures. For this purpose, this study presents an analysis of scaffold safety monitoring using strain measurement datasets. Deep neural network models were trained to learn different strain data patterns of scaffold members, corresponding to various scaffold failure cases, under active load conditions. In this study, safety conditions for a four-bay, three-story scaffold model with 20 strain values from the scaffold members were analyzed. The excellent prediction results achieved with deep learning demonstrated the superiority of this approach over previously used scaffold monitoring approaches. Using the proposed deep learning approach, safety condition of scaffolding structures can be reliably classified in semi-real time within a fraction of a second. Therefore, the approach used in this study can be reliably used to monitor scaffolds on active construction sites, and hence, ensure safer work zones for construction workers.