Effective behavior analysis of composite using homogenization and transfer learning and process optimization for injection molding based on machine learning

Abstract

Material development and process optimization are essential for product development in the manufacturing industry. Homogenization theory can efficiently predict the effective behavior of the composites using mathematical assumptions. In this thesis, two methods are proposed to predict the effective behavior of composites. First, an adaptive incrementally affine method improved from the previous homogenization method is proposed and verified for viscoelastic-viscoplastic composites. Second, a methodology that can efficiently train a deep neural network model predicting the effective behavior of composites is proposed by combining homogenization theory and transfer learning. On the other hand, process optimization is essential to improve productivity and quality of products. In particular, injection molding is the most commonly used process for plastic products. In this thesis, frameworks for optimizing the injection molding process are proposed using multiobjective Bayesian optimization and constrained generative inverse design networks.