An ANN-based concurrent multiscale damage evolution model for hierarchical fiber-reinforced composites

Abstract

In this paper, an ANN-based concurrent multiscale damage evolution model is proposed, which is able to investigate the complex failure behaviors of hierarchical fiber-reinforced composites. In the framework of the proposed model, yarn damage evolution laws at the mesoscale are indirectly derived from the microscale representative volume element (RVE), using artificial neural networks (ANNs) as a surrogate model. A homogenized characterization method is proposed to derive the homogenized damage variables. The homogenized strain and damage variables of the microscale RVE are taken as inputs and outputs in ANNs, respectively. The dataset is generated by combining clustering with the finite element simulation. A typical kind of plain-woven composite is adopted as a benchmark material for numerical implementation and experimental verification. The numerical predictions, including the tensile properties and damage evolution, are consistent with the results from quasi-static tension experiments.