Prediction of Temperature and Moisture Concentration in Autoclave-Cured Epoxy Resin Using Physics-Informed Neural Networks

Epoxy-based composites used in the aerospace industry are highly sensitive to moisture absorption, which can lead to porosity formation during the curing process and compromise structural integrity. Therefore, accurate prediction of temperature fields, degree of cure, and moisture concentration is essential for process optimization and defect mitigation. However, classical numerical approaches for solving the coupled governing equations are computationally expensive, limiting their applicability in real-time analyses and optimization strategies. In this work, Physics-Informed Neural Networks (PINNs) are investigated for predicting the transient thermal behavior, cure kinetics, and moisture concentration in an epoxy composite laminate during autoclave curing. Two PINNs are developed: the first solves the coupled transient heat transfer and cure kinetics equations in a compositetooling system, while the second predicts the moisture concentration field in the laminate using the temperature information provided by the first network. Different network architectures are evaluated, and their performance is compared with numerical solutions obtained via the Finite Volume and Finite Element Methods. The results demonstrate that PINNs accurately reproduce temperature profiles, degree of cure, and moisture concentration, achieving high coefficients of determination, while also providing significant computational efficiency advantages during the prediction stage. These findings highlight the potential of PINNs as a robust and efficient tool for modeling complex coupled phenomena in composite manufacturing processes.