| Abstract Scope |
Accurate prediction of grain morphology and microstructural evolution is crucial for controlling process–structure–property relationships in metal additive manufacturing. This work introduces a Physics-Informed Neural–Cellular (PINC) framework, a hybrid approach combining thermodynamics-based Physics-Informed Neural Networks (PINNs) with a cellular architecture to simultaneously predict grain nucleation, growth, orientation evolution, and phase percentage evolution under AM-specific thermal histories. The simulation domain is voxelized at grain-scale resolution, where each voxel hosts an individualized PINN trained on CALPHAD and Continuous Cooling Transformation data. Inter-voxel communication is achieved through two dynamic matrices: a geometry interaction matrix encoding anisotropic neighborhood connectivity and a physics transfer matrix carrying evolving microstructural attributes, including phase fraction, orientation, and stored energy. This matrix-based communication enables adaptive, history-dependent modeling, surpassing empirical cellular automata. The PINC framework captures columnar-to-equiaxed transitions, texture formation, and grain growth phenomena, representing a significant step toward computational microstructural engineering automation and digital twin development in metal AM. |