Machine learning-based prediction and classification of hygrothermal degradation in fibre-reinforced composites across multiple architectures and ageing protocols

Abstract

This study investigated the prediction of water uptake and mechanical performance of fibre-reinforced composites under hygrothermal aging using machine learning. Eight models (ANN, BPNN, SVR, RF, RT, BRT, XGBoost, and GPR) were applied for water absorption prediction, while six datasets were used for mechanical property prediction under varying aging and manufacturing conditions. Key factors such as fiber orientation, pattern number, curing, and aging temperature strongly influenced the results. Models like GPR, SVR, XGBoost, and BRT achieved R2 > 0.95 and relative errors below 5%, outperforming classical Fick diffusion models. RF and ANN accurately predicted tensile, compressive, and hoop strength, deflection, and load–displacement behavior. These findings demonstrate that machine learning effectively captures complex relationships in composites under hygrothermal aging, providing a reliable tool for material design and durability assessment, though generalization is limited by the size and diversity of the datasets.