Virtual Experiment‐Driven Neural Network‐Based Study of Resin Curing Kinetics

ABSTRACT Thermosets are widely used in various industries because of their excellent adhesion, corrosion resistance, and high strength. The study of resin curing kinetics is crucial for process optimization because it can shorten production cycles and improve product quality. However, traditional methods such as differential scanning calorimetry (DSC) require expensive instruments and complex maintenance and operations. To address these issues, this study proposes a novel approach for studying resin curing kinetics based on a convolutional neural network (CNN) driven by virtual experiments. First, a high‐fidelity finite element (FE) model validated through experiments is developed to simulate changes in temperature and degree of cure (DoC) during the resin curing process. Next, the curing kinetic parameters in the FE model are adjusted to mimic the virtual curing experiments of different resins, generating a dataset for training the CNN model. The trained CNN model can predict the DoC at a typical point according to temperature data obtained from simple curing exothermic experiments. The accuracy and generalization performance of the trained CNN model are tested using an actual resin not included in the dataset, indicating that the predictions of the CNN model deviate from the DSC results by < 5%.