Moisture absorption behavior of CFRP: Finite element analysis and machine learning coupled method

Abstract

In recent years, carbon fiber reinforced polymer (CFRP) has been widely used in the lightweight design of structures. With the increasing frequency of acid rain, the service environment of CFRP has become more acidic. Furthermore, compared to metals, CFRP is more susceptible to degradation in acidic conditions. Therefore, this study combines finite element (FE) analysis with machine learning techniques to investigate the moisture absorption characteristics of CFRP in acidic environments. First, acid-thermal aging experiments were conducted, and the moisture absorption data were fitted using the three-dimensional Fick law. These parameters were then employed to establish an FE model of CFRP moisture absorption in ABAQUS. Although this model slightly underestimates the saturation moisture absorption rate due to the omission of edge effect, it still accurately reflects the actual moisture absorption process of CFRP. Subsequently, the moisture absorption behavior of CFRP with varying thicknesses was analyzed, revealing that thinner CFRP specimens absorb moisture more rapidly, while thicker specimens exhibit slower absorption rates. Additionally, it was found that larger specific surface areas can induce edge effect, increasing the moisture diffusion coefficient in thickness direction and consequently raising the overall saturation moisture absorption rate. Finally, a CNN-LSTM deep learning model was developed to predict the moisture absorption behavior of CFRP. CNN-LSTM model demonstrated excellent predictive accuracy with significantly lower mean squared error (MSE = 8.48 × 10⁻⁷) compared to traditional models (Langmuir: 1.09 × 10⁻⁵, FE: 5.22 × 10⁻⁶), showing its superior capability in capturing complex nonlinear behavior and strong generalization potential. This study provides a novel approach for assessing the service life of CFRP in complex environments.