Machine learning-based investigation of δ phase precipitation behavior in IN718 alloy and its influence on tensile properties

Zhixuan Cheng, Jianli Zhou, Guanghao Guo, Shuang Chen, Xianjue Ye, Xiao Wei, Yuefei Zhang, Ze Zhang,
Machine learning-based investigation of δ phase precipitation behavior in IN718 alloy and its influence on tensile properties,
Materials Science and Engineering: A,
Volume 945,
2025,
148977,
ISSN 0921-5093,
https://doi.org/10.1016/j.msea.2025.148977.
(https://www.sciencedirect.com/science/article/pii/S0921509325012018)
Abstract: Traditional experimental methods have limited the comprehensive understanding of how the δ phase affects the tensile properties of IN718 alloy. This study proposes a machine learning-based approach to systematically investigate the influence of the δ phase on tensile performance. The TernausNet-16 model was used in transfer learning on 1596 scanning electron microscope images, achieving Dice and IoU scores of 75.1 % and 80.6 %, respectively. It enabled automated extraction of the δ phase morphology and content from 36 groups of aged samples. Then a random forest regression model was developed to predict microstructural features and tensile properties with R2 values exceeding 97 %. Shapley Additive Explanations (SHAP) analysis revealed the nuanced effects of the δ phase on tensile performance: (1) A small amount of δ phase at grain boundaries improves strength by hindering grain boundary motion, whereas excessive needle-like δ phase within and along grain boundaries enhances strength via the Orowan mechanism but reduces ductility. (2) The effect of the δ phase on tensile properties is nonlinear, influenced by its precipitation behavior and competition with the γ'' phase. This work demonstrates a novel strategy for microstructure-performance correlation analysis and highlights the potential of machine learning in materials characterization and property prediction.
Keywords: Machine learning; IN718 alloy; δ phase; Image segmentation; Property prediction