Investigation of failure in anisotropic composite structures via an efficient data-driven multiscale strategy

This article delves into the investigation of failure in anisotropic composite structures through an efficient data-driven multiscale strategy. The study highlights the increasing use of heterogeneous materials in various engineering fields and the challenges posed by their complex microstructures. It introduces a novel multiscale approach that combines a parametrically homogenized continuum damage mechanics (PHCDM) framework with a deep neural network (DNN) for calibrating the surrogate damage model. The strategy involves two main computational stages: the offline stage, where a discrete state database is built through nonlinear micromechanical analyses, and the online stage, where macroscale finite element computations are performed using the calibrated surrogate model. The article presents a detailed theoretical framework, including microscopic and macroscopic modeling, and the micro-to-macro connection for homogenization purposes. Numerical examples validate the proposed approach, demonstrating its accuracy and reliability in predicting anisotropic damage evolution in composite structures. The study concludes by discussing the advantages of the proposed multiscale model, such as its great numerical accuracy, excellent computational efficiency, and capability to handle nonproportional macrostrain histories. This comprehensive investigation offers valuable insights into the efficient simulation of failure in anisotropic composite structures, making it a crucial resource for professionals in materials science and engineering.