Prediction of ideal strength by machine learning

Abstract

Ideal strength is the highest stress a perfect material can withstand. In practice, a real material has a much lower strength than its ideal strength due to the existence of various defects. Knowing the ideal strength of a material is important because it can not only help to assess the gap remaining to the upper limit of the material's strength but also provides a reference for studying the fracture and deformation behavior. In this work, we developed a machine learning model to predict the ideal strength of materials. The mean R2 score of the 10-fold cross-validation for the model was estimated to be as high as 0.894, suggesting its high accuracy. In addition, a comparison between the machine learning predicted results with previously published DFT results also demonstrates the reliability. The feature importance analysis of the machine learning model unveils the mechanism that dominates the ideal strength of different materials. Based on the quantitative study between the ideal strength and the properties of the materials, novel empirical models were proposed to predict the ideal strength, i.e., the ideal strength can be quickly estimated to be 1/50 Tm or 3 ρ (density) in the unit of GPa.