Uncertainty quantification of the dimensional variations of a curved composite flange

Composite manufacturing involves a large number of materials data and process parameters, inevitably resulting in performance variations of final parts. To maintain and improve product quality, characterizing the stochastic performance of composite manufacturing processes and thus identifying the key uncertainty inputs becomes critical. However, the conventional way to quantify the stochastic behavior of composite manufacturing is primarily based on Monte Carlo simulations, which are computationally prohibitive because of complex finite element (FE) process models. In this study, we employ the Gaussian process (GP) technique to investigate the variations of the spring-in behavior of a curved composite flange fabricated using the resin transfer molding (RTM). The data-based GP technique, which is commonly used in the field of machine learning and data processing, is employed to emulate the numerical sampling required in evaluating the spring-in angle of a curved composite part. The fundamental theory behind the GP is extended from the multivariate Gaussian distribution on a finite-dimensional space to a random function defined on an infinite-dimensional space. When interpreted from a Bayesian perspective, the GP technique becomes a powerful tool to emulate complex simulations, such as advanced manufacturing processes. The effectiveness of the proposed uncertainty quantification (UQ) technique is demonstrated by predicting the processing-induced variations of the spring-in angle and identifying the key material properties that affect the prediction.