Evaluating the efficacy of singular spectrum transformation in detecting working posture changes in a time series

Abstract

In the current workplace, most manual labor is composed of high-frequency tasks with low physical workloads. Moreover, traditional ergonomic evaluation methods often have difficulty identifying slight variations in working postures and physical workloads in manual tasks. The aim of this study is to determine whether singular spectrum transformation (SST) can detect changes in human posture during manual tasks. In an experiment, eleven male participants performed lightweight material handling tasks under differing work conditions and task intervals, and an electromagnetic motion-tracking system measured their working postures. An anomaly score for each joint angle was calculated using SST, and the means, coefficients of variation (CV), and over-threshold values recorded during each experimental condition were compared. Lag is an important SST parameter for detecting how working posture differs between tasks. Therefore, the effects of changes in lag on the anomaly score were investigated. For each joint angle, the mean anomaly scores were greater under random task intervals than under constant intervals. In contrast, the CV of the anomaly score was smaller under random intervals than under constant intervals. The number of over-threshold values was significantly larger under random intervals than under constant intervals when SST was applied to the elbow flexion angle. The lag was determined according to the time of the work cycle and agreed with lag times observed in previous studies. This study concludes that the efficacy of SST was shown through detection of working posture changes in a time series, and that lag should be selected in accordance with the work cycle.

Keywords

• anomaly detection
• singular spectrum transformation
• working posture measurement
• work analysis
• human motion analysis
• movement variability