Thermo-mechanical multiphysical properties prediction of fiber-reinforced composites based on micromechanics and machine learning

The various physical properties of unidirectional fiber-reinforced composites are associated to differing extents with the performance of their constituent materials and the distribution of the fibers. To this end, this study establishes a CNN-ANN model that simultaneously considers both image data and text data, leveraging Convolutional Neural Networks (CNN) and Artificial Neural Networks (ANN). Representative Volume Elements (RVEs) were generated using a Greedy-based Generation (GBG) algorithm, and subsequently, a dataset comprising 1400 images representing different fiber distributions, along with their associated fiber nearest neighbor text data, was constructed through mesoscale finite element analysis. The fitting degree R² of the proposed CNN-ANN fusion model to the thermal conductivity of unidirectional flax fiber reinforced polymer composites (FFRPCs) exceeds \text99\% . The fit for the transverse tensile modulus, transverse tensile strength, and initial failure strain reaches R² values of \text99.7\% , \text94.9\% , and \text87.1\% , respectively. Compared to the ANN model based solely on text data, the CNN-ANN fusion model demonstrates significantly higher predictive accuracy. The findings of this study validate the feasibility and superiority of the CNN-ANN fusion model that integrates image and text data, providing valuable reference for the future application and research of machine learning methods in the prediction of the multiphysical properties of composite materials.