Process-Driven Input Profiling for Plastics Processing

Most plastic processing set points are variables that need to be defined for each sample point of the cycle. However, in the absence of on-line measures of part quality, the set points cannot be defined by feedback and need to be prescribed a priori for the entire cycle. In conventional practice, the shape of each set-point profile is defined either heuristically, according to qualitative experience, or mechanistically, to enforce a predefined trajectory for a simulated internal process state that is used as a surrogate measure of part quality (e.g., the velocity profile defined to dictate a constant melt front velocity). The purpose of this study is twofold: (i) to evaluate the efficacy of using a single internal state as the surrogate of part quality, and (ii) to explore the feasibility of devising a multivariate profiling approach, where indices of multiple process states act as surrogates of part quality. For this study, an injection-compression molding process used for production of digital video disks was considered as the development domain, and a pseudo-optimal cycle of the process was found by reinforcement learning to provide a basis for evaluating the ideal behavior of the process states. Compared to conventional molding, the results indicate that the asymmetric process capability index, CPK, was increased by $∼50%$ with velocity profile optimization and to 120% with both velocity profile and pressure profile optimization. Two general conclusions result. First, velocity and pressure profiling provide important degrees of freedom for optimizing process control and maximizing part quality. Second, estimators for unobservable process states, in this case birefringence and warpage, can be used to develop different machine profiles to selectively trade off multiple quality attributes according to user preferences.