Evaluation for Workability Based on Anomaly Detection using One-Class Support Vector Machine

: Application in Screw Tightening Task Near the Body

Abstract

Extracting the working posture of different trend （anomaly posture） using anomaly detection helps to understand the problems of posture in the work tasks. This study proposes the extraction method for anomaly posture using One-Class Support Vector Machine. The proposed method was applied to the screw tightening tasks in various working positions and heights near the body on twelve participants. During the tasks, Euler angle of each body segment, joint angle and calculated joint torque ratio were recorded, and the participants were questioned about their subjective evaluation. The proposed method calculated the ratio of extracted anomaly posture under several settings of the hyperparameter and the specific vector. In the near and central work position, subjective difficulty of the work increases while does not show the relationship to the joint angle and joint torque ratio. On the contrary, the proposed method shows the relationship of the ratio of extracted anomaly posture to subjective difficulty when the hyperparameter ν, which is upper bound on outlier, set to 0.10.