Scaffold Safety Analysis: Focusing On Divide-And-Conquer Method

Researchers in construction safety have invested a significant amount of time on improving and automating the safety of scaffolding structures. However, the developed applications are limited to identifying a small number of failure cases for scaffolds, leaving their applicability for large classification problems uncertain. Such uncertainty necessitates further investigation that can broaden these applications to make them capable of identifying a wider range of scaffold failure cases. One challenge in a large classification problem is associated with the accuracy and computational demand required for making safety predictions. To mitigate these two aspects, this study presents an approach using the divide-and-conquer method for monitoring scaffolding safety. As a classification method, this study uses machine learning, while particular interest is placed on the effect of applying a divide-and-conquer technique to accuracy and prediction time, which are particularly important in real-time applications. The approach divides the overall scaffold-safety classification problem into subsets of smaller problems in a hierarchical structure, and classification is then conducted on the subsets of the problems. Test results showed that this approach could classify 1,411 failure modes of a four-bay, three-story scaffold by using only 20 strain measurement values from 97% to 100% with an average of 99% accuracy within 0.0065 seconds of average prediction time, after training. This study also confirms the applicability of the divide-and-conquer method for facilitating large classification problems more effectively than conventional approaches.