Characterization and property prediction of fibre structures within discontinuous-fibre reinforced polymer matrix composites using 3D fibre cells assisted by contrastive learning

Abstract

Fibre-cell-based fibre structure characterization approach was proposed recently to characterize the fibre distribution within discontinuous-fibre reinforced polymer matrix composites (DFR PMCs) over a 2D domain. This approach determines the distribution state of each fibre based on the relative size and topological features of its fibre cell. In this study, the fibre-cell-based approach is extended for 3D fibre domains. A convolutional neural network (CNN) encoder is trained through contrastive learning to quantitatively represent topological features of 3D fibre cells. Subsequently, the feature–property correlations are established using an artificial neural network (ANN). For practical application, the ANN is integrated with an image analysis software to provide in situ predictions of local elastic modulus of a DFR PMC based on its fibre structures observed from micro-CT images. The predictions are also compared with the experimental measurements acquired through microindentation testing, and it shows a good agreement.