Novel Unsupervised Machine Learning Method for Identifying Falling from Height Hazards in Building Information Models through Path Simulation Sampling

Abstract

Falling from height (FFH) is a significant safety concern in the construction industry. It requires construction safety managers to identify and prevent FFH from occurring as early as possible. However, due to insufficient expertise and the shortage of relevant personnel, it is often challenging to accurately and promptly identify FFH risks, especially those more dangerous ones. To address this issue, we proposed a new identification method for FFH risk points by incorporating BIM, path simulation, and machine learning techniques. Our approach, based on the likelihood exposure consequence criterion, efficiently identified and prioritized hazardous FFH risk points. Unlike prior studies, we emphasized the spatial distribution of workers by incorporating site layout and construction schedule considerations. The algorithm generated a safe routing plan, which was validated in experiments, emphasizing its effectiveness in early detection and mitigation of FFH risks. This research provided a comprehensive approach to FFH risk management that integrates building information modeling, path simulation, and machine learning to comprehensively address FFH risks in construction and generate safe route plans for effective safety management. The proposed method significantly contributes to the early detection and elimination of potential FFH risks in construction projects.