Nonlinear and damage micromechanical surrogate models for multiscale simulation of compression-after-impact in composite laminates

Predicting the initiation and progression of internal damage in composite laminates remains a major challenge in structural design. Physics-based micromechanical models can represent complex material behavior and damage mechanisms in detail; however, their direct integration into structural simulations is computationally prohibitive. This study introduces a data-driven surrogate model based on artificial neural networks (ANNs), trained using Parametric High-Fidelity Generalized Method of Cells (PHFGMC) micromechanical simulations, to predict nonlinear response and damage evolution in fiber-reinforced polymer composites under multiaxial loading. A key contribution is the introduction of reduced-history parameters- defined as the strain corresponding to the maximum or minimum stress reached in each loading mode- which enable the ANN to capture path-dependent matrix and fiber damage without recurrent architectures or extensive stored sequences. Calibrated for a carbon/epoxy material system, the surrogate model is applied to low-velocity impact (LVI) and compression-after-impact (CAI) simulations of laminated plates, with interlaminar delamination captured through cohesive surfaces. Predictions are validated against experimental force–displacement responses, ultrasonic C-scans, micro-CT imaging, and digital image correlation measurements. The results demonstrate that the proposed framework effectively bridges micromechanical fidelity with structural-scale applicability, enabling physically informed damage predictions for composite structural analysis.