In this study, we propose to explore inverse optimization process to better understand human arm motion in industrial screwing task. The process combines several criteria to minimize such as energy expenditure or trajectory smoothness leading to the optimal trajectory of a typical screwing task, often performed by workers. Estimated joint trajectories are similar with the measured ones, with a mean square error of 4 degrees. The resulting cost-function is mainly composed of energy expenditure and geodesic criteria. Results show the relevance of using composite cost function in human motion planning. This study has been conducted to assist workers by using collaborative robots in painful task in PSA Peugeot Citroen factories to improve ergonomics of manual workstations.