Atomistic-Informed and Machine Learning–Assisted Crystal Plasticity Modeling for Material Interfaces

This paper presents a novel methodology that addresses the limited predictive capability of the existing crystal plasticity (CP) method in interfaces modeling. Our approach incorporates interfacial parameters generated from molecular dynamic (MD) simulations into the continuum-level crystal plasticity finite element analysis (CPFEA) model. To address the inherent scale mismatch between atomistic and continuum models, we employ two essential techniques—the nucleation theory and machine learning (ML) method. The nucleation theory is utilized in conjunction with the nudged elastic band (NEB) method to extrapolate the low strain-rate yield stresses from the high strain-rate MD simulation results. To overcome the length scale limitation of MD, we use a method that was recently developed by the authors—a two-step approach that utilizes MD-calculated stress–strain data to train a probabilistic ML model for predicting stress–strain behaviors at larger scale. The resulting flow parameters and extrapolated yield stresses are then integrated into the atomistic-informed interface region of the CPFEA model. This multiscale computational method that combines MD, CPFEA, nucleation theory, NEB and ML enables a grain-level large time scale crystal plasticity modeling with atomic accuracy at the interfaces, as demonstrated by carefully validating it through a bicrystal Cu model with experimental results. This validation highlights the importance of accurately describing interfaces in the modeling of material mechanical behavior. Notably, our proposed methodology is not limited to interfaces but can be applied to other microstructures requiring atomic accuracy. The method opens up new possibilities for comprehensively understanding and designing materials with complex microstructure for various engineering applications.