Applications of Machine Learning for Process Modeling of Composites

Abstract

Science-based simulation tools such as Finite Element (FE) models are widely used in engineering applications including process modeling of composites. There are inherent limitations associated with these models including trade-off between fidelity and cost, inability to tackle uncertainties such as unknown manufacturing boundary conditions, and difficulties with inverse modeling and optimization of multi-dimensional problems. With the rise of Machine Learning (ML) and data-driven modeling, many branches of science and engineering are exploring the applications of these methods with varying degrees of success. Here we explore current applications of ML in process simulation of composites. With several case studies, it is demonstrated how some of the limitations of traditional science-based models can be addressed using a combined FE-ML approach in the new paradigm of Theory-Guided Machine Learning (TGML). Specifically, case studies are presented for thermo-chemical analysis of composites processing, where surrogate Neural Networks (NN) are developed for near real-time modeling of the manufacturing process.