Abstract
Predicting the corrosion fatigue life of austenitic stainless steel (AusSS) in high-temperature, high-pressure water environments remains a complex task due to the intricate interaction of thermal, mechanical, and environmental factors, along with the scarcity of comprehensive experimental data. Traditional empirical fatigue models often fail to capture these multifactorial effects, limiting their predictive accuracy. This study presents a novel machine learning framework using a Genetic Algorithm Optimized Neural Network (GAONN) to enhance corrosion fatigue life prediction. The proposed model is applied across five AusSS grades, 316, 316LN, 316NG, 304, and 308L, offering generalized predictions across different alloy compositions and operational scenarios. Key input parameters include environmental factors such as temperature, pressure, and dissolved oxygen (DO), mechanical properties like strain rate and amplitude, and metallurgical attributes such as stacking fault energy (SFE). The GAONN model demonstrates excellent predictive accuracy on the test set, achieving an R(2) value of 94.4%, a mean-squared error of 0.014, a root mean-squared error of 0.120, a mean absolute error of 0.105, and a mean absolute percentage error of 3.6%. These results represent a significant improvement over classical corrosion fatigue life models. Additionally, Shapley Additive Explanations (SHAP) analysis identifies temperature, pressure, and strain amplitude as primary contributors to corrosion fatigue life, in agreement with metallurgical theory. This work establishes GAONN as a transparent, accurate, and scalable tool for assessing corrosion fatigue life in next-generation nuclear energy systems.