An Accelerated Process Optimization Method to Minimize Deformations in Composites Using Theory-guided Probabilistic Machine Learning

Abstract

While aerospace manufacturing and assembly technologies have significantly evolved, challenges persist in mitigating process-induced deformations (PIDs) in composite parts. Physics-based-simulation optimization strategies have been developed to address these challenges. However, they often prove ineffective or inefficient because of the time-consuming, costly, and deterministic characterization efforts they require. This study introduces a novel characterization- and simulation-free optimization approach for predicting and minimizing PIDs in composite parts. The method uses a theory-guided machine learning (TGML) framework, consisting of limited element-level experiments and Gaussian Process Regression (GPR) with integrated closed-form domain knowledge. This paper showcases the effectiveness of the proposed TGML method for two case studies, where an optimal layup and cure cycle are found and validated after a mere six experiments and without any material characterization or simulation efforts. The method presented in this paper aims to offer a cost-efficient and generalizable process optimization procedure to potentially mitigate PIDs in composite parts.