Optimization of Variable Blank Holding Force Conditions for Improvement of Formability in the Cylindrical Cup Drawing Process Using a Reinforcement Learning Algorithm

In this study, the variable blank holding force (VBHF) technique is applied to the cup drawing process. To increase and optimize formability, a reinforcement learning (RL) algorithm was employed to determine the VBHF trajectory with respect to the punch stroke. A Markov decision process (MDP) was constructed to formulate the trajectory optimization problem within the RL framework. A finite element (FE) simulation model of the drawing process was developed and integrated with the algorithm. The reward signal was defined by considering the defects such as wrinkling and fracture of the sheet, which frequently occur in the deep drawing process. The objective of this study was to identify the optimal control strategy for the VBHF, maximizing both the limit drawing height (LDH) and the minimum thickness of the sheet. The results demonstrate that the proposed algorithm can effectively discover a control strategy that enhances the formability of the sheet.