A self-supervised learning framework based on physics-informed and convolutional neural networks to identify local anisotropic permeability tensor from textiles 2D images for filling pattern prediction

In liquid composite molding processes, variabilities in material and process conditions can lead to distorted flow patterns during filling. These distortions appear not only within the same part but also from one part to another. Notably, minor deviations in the dry fibrous textiles cause local permeability changes, resulting in flow distortions and potential defects. Traditional permeability models fall short in predicting these localized fluctuations, especially for anisotropic textiles, whereas reliance on homogeneous permeability models creates substantial discrepancies between forecasted and observed filling patterns. This study presents a self-supervised framework that determines in-plane permeability tensor field of textiles from an image of that textile in dry state. Data from central injection experiments is used for training, including flow images and pressure inlet data. This work demonstrates that this model proficiently predicts flow patterns in unobserved experiments and captures local flow distortions, even when trained on a relatively small dataset of experiments.