A data-driven low-cycle fatigue life prediction model for nickel-based superalloys

Abstract

A data-driven machine learning method for the low-cycle fatigue (LCF) life prediction of Nickel-based superalloys is proposed to overcome the limitations of empirical formulas. The method integrates three characteristics of the superalloys: chemical composition, heat treatment process and experimental parameters, and constructs a unified fatigue life prediction dataset. The relationship between many different microstructural parameters and LCF life is analyzed using the Pearson Correlation Coefficient (PCC) and Maximal Information Coefficient (MIC). The fatigue life dataset is then used to predict the LCF life with four machine-learning models: Random Forest (RF), Artificial Neural Network (ANN), Support Vector Machine (SVM), and a Genetic Algorithm-based Random Forest (GA-RF). The comparative study results demonstrate that the models accurately predict the LCF life, as evidenced by the respective R2 values of 0.8311, 0.8345, 0.7711, and 0.9272, with GA-RF performing the best and equally better in comparison with the Coffin-Manson model. The proposed method efficiently maps feature-life relationships for the LCF life of Nickel-based superalloys, reducing experimental time and cost and promising applications in the inverse design and manufacture of alloys.