Machine learning-based screening of high-entropy alloy AlCoCrFeNi particles produced via high-energy mechanical alloying method

Abstract

High entropy alloys (HEAs) are widely explored due to their superior material properties. However, while mechanical alloying is a simple and cost-effective laboratory route for producing HEA particles, localized energy concentrations often induce segregation, and current alloying assessment methods, such as EDS, are time-consuming and can yield misleading compositional maps. This study proposes a deep learning regression approach to predict compositional homogeneity, distinguishing alloyed/segregated particles directly from SEM images. The approach utilizes compositional uniformity derived from per-particle EDS composition vectors as ground truth for the AlCoCrFeNi system. Regression models were trained to directly predict compositional uniformity via Shannon entropy (H). DenseNet121 achieved an R² of 0.918 and an MAE of 0.031, corresponding to an average deviation of ~3% of the full alloying uniformity scale. This approach enables rapid, automated screening of HEA powder and can be integrated into existing laboratory workflows for quality assessment of AlCoCrFeNi systems using SEM alone.