Developing Joint-level Scoring Models Tailored to Whole-body Ergonomic Assessment of Construction Workers

Construction workers frequently engage in manual operations at workplaces, increasing their ergonomic risks of developing Work-Related Musculoskeletal Disorders (WMSDs). To proactively assess and prevent such risks, ergonomic scales have been widely employed, incorporated cutting-edge technologies to achieve advancements in automation. However, these scales have demonstrated limited accuracy in risk identification, mainly attributed to unreliable joint-level assessment rules based on discrete boundaries and binary rules. Although previous attempts have incorporated fuzzy logic to improve accuracy, the involved subjective determination of function shapes and threshold settings remains a persistent hindrance. To address this limitation, the present study aims to develop data-driven joint-level scoring models for replacing these conventional rules. This process leverages pose data from the Construction Motion Data Library (CML) dataset and employs a robust and heuristic data-driven approach named Heuristics Gaussian Cloud Transformation (H-GCT). The results, with all Confusion degrees below the threshold value of 0.64, demonstrate the significant independence of the developed scoring models, ensuring accurate identification of ergonomic risk. Furthermore, a comparison is conducted with previous studies that employed fuzzy logic to improve REBA. This process highlights the superiority of the data-driven H-GCT in developing scoring models. This study contributes to the existing body of knowledge by providing joint-level scoring models to improve the applicability of ergonomic assessment in construction. Future studies can further enhance this work by expanding pose data, enriching assessment modules, and refining the data-driven approach.