Abstract
Carbon fiber reinforced polymer (CFRP) components require precise curing process control to ensure quality, but traditional phenomenological cure kinetics models face limitations in handling nonlinearity and data diversity. This study addresses the challenges in modeling the cure kinetics of carbon fiber reinforced polymer (CFRP) composites, where traditional phenomenological models lack generalizability and neural networks suffer from robustness issues due to their numerous hyperparameters and data dependency. To overcome these limitations, a novel machine learning model called the angle information-enhanced radial basis function (RBF) model is proposed, which integrates both Euclidean distance and angular relationships between data points to improve prediction stability and accuracy. The performance of this machine learning approach is systematically compared against an autocatalytic model and a neural network using dynamic DSC data from T700/2626 epoxy resin at multiple heating rates. The angle-enhanced RBF model balances accuracy, efficiency, and robustness, offering a reliable data-driven alternative for CFRP cure kinetics prediction without requiring extensive data or complex hyperparameter optimization, thus facilitating better process control in manufacturing.