Input Profiling for Injection Molding by Reinforcement Learning

Abstract

An adaptation method is investigated for improving the shape of input profiles in injection molding. The noted characteristic of injection molding is that performance feedback (i.e., part quality measure) becomes available only at the end of the cycle, therefore, the performance of the entire sequence of inputs that form the profile is evaluated by the same delayed measure at the end of the cycle. The proposed profiling method uses the concept of reinforcement learning, which is particularly suited to problems with delayed feedback. For an initial study, the method is tested in improving the profiles of the ram velocity and packing pressure. For this study, a simulation program is used to provide estimates of digital video disks (DVDs) quality attributes as feedback for evaluating the performance of the adapted profiles. The initial results indicate that the proposed method is effective in refining the profiles, which will lead to better quality parts with faster cycles.