작업 자세 평가를 위한 Vector-based 및 Deep Learning-based OWAS 평가 시스템 개발

Abstract

Objective: This study was intended to develop two risk assessment systems that can evaluate OWAS scores and action level with good accuracy and reliability to prevent work-related musculoskeletal disorders (WMSDs). Background: Since risk assessment methods such as OWAS and REBA are relied on visual observations, their accuracy and reliability might have been in questions. Therefore, computer-aided risk assessment systems are necessary to enhance accuracy and reliability as well as to improve applicability in workplaces. Method: This study developed vector-based OWAS (V-OWAS) and deep learningbased OWAS (DL-OWAS). V-OWAS (partial automation) asks a user to manually identify the key joint positions from a working picture and automatically calculates OWAS scores and action level. DL-OWAS (high automation) automatically identifies the key positions and determines OWAS scores and action level. To evaluate the performance of the two systems, we used 40 working pictures and compared the results of the two systems with the results of four ergonomic experts with over 10- year experience who did assess based on traditional visual observations. Results: The two systems showed similar assessment results with those obtained from the ergonomic experts. The agreement percentage of V-OWAS (91.3%) with the experts was slightly better than DL-OWAS (90%). On the other hand, the usability score of DL-OWAS (7 out of 7 point Likert scale) was significantly higher than V-OWAS (5.75) and traditional visual observation (5.25). Conclusion: V-OWAS had an advantage in accuracy of identifying the key joint positions; but it required more time and efforts over DL-OWAS. DL-OWAS can assess automatically; but its accuracy might be low if some parts of the body are obscured due to some reasons (e.g., obscured by some machine). Therefore, it is recommended to use V-OWAS or DL-OWAS according to the condition of working picture. Application: The two OWAS systems developed in this study can help industrial practitioners effectively and efficiently assess working posture for preventing WMSDs.