Machine learning-assisted creep life prediction and empirical formula generation for 9-12% Cr steel

Abstract

The creep behavior of steels is influenced by factors such as lattice structure, defects, and stress conditions. Given the high cost and time required for creep tests, accurately predicting creep life and minimum creep rate is essential. This study analyzed 9–12% chromium (Cr) steel using a dataset of 1496 entries covering material composition, mechanical properties, minimum creep rate, and creep life. Nine machine learning (ML) models were developed, with the artificial neural network (ANN) achieving the highest prediction accuracy, evidenced by a coefficient of determination (R²) of 0.9973 and a root mean square error (RMSE) of 0.045. A dual-target neural network model provided R² values of 0.9853 for creep life and 0.9838 for minimum creep rate. Additionally, an empirical equation based on gene expression programming (GEP) achieved an R² exceeding 0.9741. This study offers novel insights into the design of 9–12% Cr steel with enhanced creep life.