Integrated Proxy Micromechanical Models in Multiscale Analysis Using Deep Learning for Laminated Composites Subject to Low-Velocity Impact

Abstract

Multiscale nonlinear micromechanical approaches can analyze laminated composite structures. In contrast, current classical macromechanical modeling approaches depict composite materials as anisotropic homogenized media. This research proposes alternative refined micromechanics that can generate the local mechanical behavior of fiber and matrix constituents and accurately depict the microstructure. The parametric high-fidelity generalized method of cells (PHFGMC) is an advanced micromechanical method that can be used for the nonlinear and failure analysis of different composite material systems. The computational effort required for generating the nonlinear multiaxial behavior is relatively small, depending on the size of the discretized repeating unit cell (RUC). However, it is computationally challenging, if not impossible, to integrate refined nonlinear micromechanical models within a multiscale finite element (FE) analysis of composite structures. To that end, we propose a new artificial neural network (ANN) based micromechanical modeling framework, termed ANN-PHFGMC, for depicting the nonlinear behavior of fiber-reinforced polymeric (FRP) materials. Pre-simulated mechanical stress-strain responses and behaviors are determined using the PHFGMC to generate a multiaxial training database for the ANN micromodel. The PHFGMC effective stress-strain responses for different applied multiaxial strain paths are divided into two sets of data; one for the training and the other for verifying the trained ANN-PHFGMC model. The resulting trained ANN-PHFGMC is accurate, with less than a 5% error in the verified predictions. Next, the ANN-PHFGMC model can be integrated within a commercial explicit FE code for multiscale low-velocity impact (LVI) analysis of laminated composite plates.