| By learning the non-linear relationships that govern the injection molding process, the ANN configuration developed in this research is able to interpolate the proper injection molding machine operational settings for parts that have not been injection molded before. This learning process is based on training and testing the developed ANN with data gathered from injection molding a number of geometrically varying parts. This variation in geometry is captured through a set of unique and relevant numbers that are generated through two creative techniques developed in this dissertation. The two techniques which are combined to give the total set of geometrical features are, free space and the critical path method. The geometrical features generated from the first technique are more conventional and include volume, surface area, area projections, length, and thickness. The second technique looks at area integrals along the most critical path the injected resin will follow inside the mold cavity. In other word, the critical path method or technique attempts to look at the effects of geometry through the eyes of the resin as it fills the mold cavity.;The final data sets used in training and testing the different ANNs developed in this research include process characteristics, geometrical features, corresponding machine settings, and quality features for five geometrically varying parts. This data allows the developed ANN to learn, or more accurately map the operational space of the injection molding machine. The developed ANN is then able to interpolate the proper machine's operational settings that will generate the desired quality features for a part never before molded.;The data gathered from actual injection molding runs is transformed from its raw form to a set of process characteristics that further define the injection molding process. These process characteristics include peak, slope, and integral values measured from the different temperature and pressure sensors. These process characteristics allow the quality of an injection molded part to be predicted while the part is still in the molding machine. This quality predictive capabilities is generalized in this dissertation to work on multiple machines, multiple resins, and multiple part designs. |
