Abstract
With the vast development on computer hardware as well as computer aided engineering (CAE) software, injection molding has become one of the primary methods for producing precision industrial products. Hence, the problems on improving both production efficiency and product quality need to be immediately studied and solved today. After reviewing the literatures and theoretical background concerning the process control for injection molding in the past decades, the study on quality improvement and experimental verification has been thoroughly conducted in this dissertation. With the considerations for quality engineering methodology, various viability studies and design of experiments have been accomplished. Also, improvements on the basic process control technology have been tested based upon the theoretical approach via neural networks modeling. In this study, a novel optimization scheme, namely inverse neural optimal control system (INOCS), is proposed for the process control with two serially connected radial basis functions networks (RBFN) that act as a quality predictor and an optimal controller relying on a multi-losses function based performance index. The proposed INOCS controller can appropriately handle multi-qualities or various combinations of qualities with prescribed weightings. After being verified by numerical simulations and experiments, the injection molding process could be fully automated with processing parameters set to the optimal conditions via the minimization of the total loss index. The INOCS controller works for not only the initialization of the parameters during startup but the optimization and adaptation of parameters during the process. In conclusion, the controller should meet the quality requirements and maintain steady operating conditions in less than a few cycles.