Machine Learning Assisted Prediction of Defect Absorption Rates at Grain Boundaries Based on the Molecular Dynamics Simulation

Abstract Grain boundary (GB) plays a crucial role in the mechanical properties and irradiation resistance of nuclear materials. It is thus essential to understand and predict the defect behaviors near GBs. Here, we present a framework for predicting defect absorption rates (fα) near GBs in four face-centered cubic metallic systems (pure Cu, Cu70Co30, pure Ni, and NiCoCr) by machine learning (ML). An extensive dataset was compiled by varying the primary knock-on atom energies, GB types, and material compositions, resulting in 141 distinct molecular dynamic simulations. The key GB characteristics such as tilt angle, GB energy, and coincident site lattice were selected to construct the descriptors, supplemented by four variables related to defect formation energy to capture the thermodynamics of atomic-scale interactions. The optimal descriptors, combining both chemical and structural descriptors, were determined through the Pearson correlation analysis. Six machine learning algorithms were applied to identify the best model, with the random forest model achieving the highest cross-validated determination coefficients (R2) of 0.88 for interstitials and 0.80 for vacancies. Additionally, Shapley additive exPlanations analysis was employed to elucidate and interpret the predicted defect absorption rates from the ML models, identifying GB energy (γGB) and interaction width (dGBα) as dominant regulators. The present work establishes the relationship between the defect absorption rates and the GB structure via ML and shows great prospect in the application of ML methods on modeling GB-relevant defect properties.