Descriptive modeling of textiles using FE simulations and deep learning

In this work we propose a novel and fully automated method for extracting the yarn geometrical features in woven composites so that direct parametrization of textile reinforcement is achieved ( e.g. , FE mesh). Thus, our aim not only to perform segmentation from tomographic images but rather provide complete descriptive modeling fabric. As such, approach improves on previous methods use voxel-wise masks as intermediate representations followed by re-meshing operations (yarn envelope estimation). The proposed employs two deep neural network architectures (U-Net Mask R-CNN). First, train U-Net generate synthetic CT corresponding simulations. This allows large quantities annotated data without requiring costly manual annotations. then used R-CNN, which focused predicting contour points around each yarns image. Experimental results show accurate robust performing instance images, further validated quantitative qualitative analyses.