Karatas I. Deep learning-based system for prediction of work at height in construction site.
Heliyon 2025;
11:e41779. [PMID:
39906815 PMCID:
PMC11791131 DOI:
10.1016/j.heliyon.2025.e41779]
[Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/09/2024] [Revised: 01/06/2025] [Accepted: 01/07/2025] [Indexed: 02/06/2025] Open
Abstract
Falling from height (FFH) is a major cause of injuries and fatalities on construction sites. Research has emphasized the role of technological advances in managing FFH safety risks. In this investigation, the objective is to forecast if a laborer is operating at an elevated position by utilizing an accelerometer, gyroscope, and pressure information through the application of deep-learning techniques. The study involved analyzing worker data to quickly implement safety measures for working at heights. A total of 45 analyses were conducted using DNN, CNN, and LSTM deep-learning models, with 5 different window sizes and 3 different overlap rates. The analysis revealed that the DNN model, utilizing a 1-s window size and a 75 % overlap rate, attained an accuracy of 94.6 % with a loss of 0.1445. Conversely, the CNN model, employing a 5-s window size and a 75 % overlap rate, demonstrated an accuracy of 94.9 % with a loss of 0.1696. The results of this study address information gaps by efficiently predicting workers' working conditions at heights without the need for complex calculations. By implementing this method at construction sites, it is expected to reduce the risk of FFH and align occupational health and safety practices with technological advancements.
Collapse