RVE ‐Based Mechanical Modeling of Woven CFRPs With Nonorthogonal Yarn Angles via Micro‐ CT Scan and Deep Learning

ABSTRACT

This work mainly explores the effectiveness of using deep‐learning‐based reconstruction methods from CT scan images to directly establish the representative volume element (RVE) of woven composite materials with nonorthogonal yarn angles for high‐fidelity finite element (FE) simulation. Compared to traditional RVE generation software, using CT scan images combined with deep learning reconstruction for RVE can effectively avoid the problem of yarn‐to‐yarn penetration. Meanwhile, with the addition of the self‐developed post‐processing algorithms, the key parameters of RVE elements, including local fiber orientation and local fiber volume fraction, can be accurately estimated to improve the accuracy of finite element simulation. The high‐fidelity FE model constructed from the deep learning method was utilized to do bias‐extension simulation. Experimental validation indicates that the deep learning‐based FE framework can predict the mechanical performance of the cured woven CFRP specimens with large shear angle of 50° with average error of 3.35%, indicating the potential of this methodology to accurately predict the mechanical performance of woven composites with nonorthogonal yarn angles.