Application of Chemistry-Informed Neural Networks in Modeling Cure Kinetics of Prepreg Materials

Abstract This study addresses the challenge of accurately modeling cure kinetics for thermosetting prepreg composites, specifically the carbon-fiber reinforced epoxy material Toray T830H-3900, which is used in aerospace applications. Traditional models often require extensive chemical expertise and experimental data, involving complex overlapping reactions and diffusion effects. These models can yield inconsistent results due to variations in experimental conditions, necessitating repetitive and time-consuming tests. To overcome these challenges, a Chemistry-Informed Neural Network (CINN) model was developed and implemented. Unlike purely data-driven approaches, the CINN model incorporates chemical knowledge, enhancing prediction accuracy and robustness. This integration helps to mitigate the risk of overfitting and improves the model’s ability to generalize across different curing conditions. The CINN model was validated through a series of robustness tests, including milligram-scale Differential Scanning Calorimetry (DSC) curing tests, gram-scale thermal disturbance tests, and kilogram-scale fabrication tests. Results showed that the CINN model outperformed traditional methods, significantly reducing prediction error—from 4.7% to 1.3%—in key fabrication tests. The model also demonstrated better mathematical stability and chemical validity compared to previous neural network approaches. By effectively bridging empirical model fitting and machine learning, the CINN model offers a reliable, efficient, and accurate solution for cure kinetics prediction, accelerating material design and process optimization in composite manufacturing.