A graph neural network surrogate for 3D finite element modeling: accelerated full-field parameter identification in aluminum alloy 6DR1

This study introduces a novel three-dimensional graph neural network (GNN) based finite element (FE) surrogate model designed to emulate full-field tensile simulations of aluminum alloy 6DR1. The proposed model captures the spatial and temporal evolution of mechanical fields while drastically reducing computational cost. Compared with conventional Abaqus simulations, the GNN achieves a speed-up of nearly four orders of magnitude (≈10,000 × faster), enabling real-time parameter identification essential for digital twin applications in manufacturing. The surrogate accurately reproduces both local and global mechanical responses, including stress, strain, and force–displacement behavior, across the entire deformation history. Sensitivity analyses reveal the most influential material parameters governing the model’s predictive behavior, while uncertainty quantification defines the parameter space in which GNN predictions remain reliable. The presented framework establishes an efficient and physically consistent approach for accelerated parameter calibration and opens a pathway toward the integration of GNN-based surrogates into real-time digital twins for process monitoring and optimization.