A unified creep and fatigue life prediction approach for 316 austenitic stainless steel using machine and deep learning

Abstract

316 Austenitic stainless steel (AusSS) is extensively utilized in high-temperature industrial applications such as boiler tubes and nuclear reactor pressure vessels. These components commonly experience failure under high-temperature and high-pressure conditions, attributed to either creep or fatigue. Existing classical models for creep and fatigue life prediction focus on a singular failure mode (either creep or fatigue) and consider physical features only. This study aims to develop a unified life prediction model for both creep and fatigue phenomena. It synthesizes information from 12 additional unexplored chemical and microstructural features from the National Institute of Materials Science (NIMS), Japan database, and previously published literature. Machine learning (such as decision tree, random forest, and XGBoost) and deep learning (like deep neural network) algorithms are employed in the modeling process. The trained models have been cross-validated against unseen creep and fatigue life data, demonstrating superior prediction accuracy of 96.1% for deep neural network compared with classical models.