DATA-DRIVEN PREDICTIVE MODEL OF RESIN FILLING TIME OF COMPOSITE MOLDING PROCESS

This paper presents an application of the manufacturing constraint modeling (MCM) method we previously developed, to composite molding processes. A statistical model to predict resin filling time for given part geometry and inlet location is constructed through massive process simulations and data mining. A bitmap representation is adopted to generate massive samples of part geometries within a bounding box. The model is trained by the statistical regression based on the abstract features inspired by the underlying physics of the filling process, which dramatically enhances the model generalizability compared to the conventional surrogate models. The model is tested by in-the-bag samples and out-of bag samples with a different inlet gate location and a different bounding box. The result shows the manufacturing constraint model trained by the knowledge-inspired feature representation achieves comparable in-the-bag training errors as the surrogate models, and remarkably better out-of-bag testing results. The proposed manufacturing constraint model will be useful to enhance the manufacturability of composite structures during manual design iterations as well as computer-based optimization.