Machine learning approach to determine the decision rules in ergonomic assessment of working posture in sewing machine operators

Introduction: There are some inherent problems with the use of observation methods in the ergonomic assessment of working posture, namely the stability and precision of the measurements. This study aims to use a machine learning (ML) approach to avoid the subjectivity bias of observational methods in ergonomic assessments and further identify risk patterns for work-related musculoskeletal disorders (WMSDs) among sewing machine operators. Methods: We proposed a decision tree analysis scheme for ergonomic assessment in working postures (DTAS-EAWP). First, DTAS-EAWP used computer vision-based technology to detect the body movement angles from the on-site working videos to generate a dataset of risk scores through the criteria of Rapid Entire Body Assessment (REBA) for sewing machine operators. Second, data mining techniques (WEKA) using the C4.5 algorithm were used to construct a representative decision tree (RDT) with paths of various risk levels, and attribute importance analysis was performed to determine the critical body segments for WMSDs. Results: DTAS-EAWP was able to recognize 11,211 samples of continuous working postures in sewing machine operation and calculate the corresponding final REBA scores. A total of 13 decision rules were constructed in the RDT, with over 95% prediction accuracy and 83% path coverage, to depict the possible risk tendency in the working postures. Through RDT and attribute importance analysis, it was identified that the lower arm and the upper arms exhibited as critical segments that significantly increased the risk levels for WMSDs. Conclusions: This study demonstrates that ML approach with computer vision-based estimation and DT analysis are feasible for comprehensively exploring the decision rules in ergonomic assessment of working postures for risk prediction of WMSDs in sewing machine operators. Practical Applications: This DTAS-EAWP can be applied in manufacturing industries to automatically analyze working postures and identify risk patterns of WMSDs, leading to the development of effectively preventive interventions.