Abstract
Thermal and nuclear power systems require materials capable of sustaining high mechanical and thermal loads over prolonged service durations. Among these, 9Cr heat-resistant steels are particularly attractive due to their superior mechanical strength and extended creep rupture life, making them suitable for extreme environments. In this study, multiple machine learning models were explored to predict the creep rupture life of 9Cr heat-resistant steels. A comprehensive dataset of 913 samples, compiled from experimental results and literature, included eight input variables-covering chemical composition, stress, and temperature-and one output variable, the creep rupture life. The optimized artificial neural network (ANN) model achieved the highest predictive accuracy with a regularization coefficient of 0.01, 10,000 training iterations, and five hidden layers with 30 neurons per layer, attaining an R(2) of 0.9718 for the test dataset. Beyond accurate prediction, single- and two-variable sensitivity analyses were used to elucidate statistically meaningful trends and interactions among the input parameters governing creep rupture life. The analyses indicated that among all variables, test conditions-particularly the test temperature-exert a pronounced negative effect on creep life, significantly reducing durability at elevated temperatures. Additionally, an optimization module enables identification of input conditions to achieve desired creep life, while the Index of Relative Importance (I(RI)) and quantitative effect analysis enhance interpretability. This framework represents a robust and reliable tool for long-term creep life assessment and the design of 9Cr steels for high-temperature applications.