A deep learning based life prediction method for components under creep, fatigue and creep-fatigue conditions.

• A general machine learning life prediction method is proposed for creep, fatigue and creep-fatigue conditions. • Creep, fatigue and creep-fatigue data are integrated into a unified dataset. • DNN exhibits better prediction accuracy than conventional machine learning models. Deep learning is a particular kind of machine learning, which achieves great power and flexibility by a nested hierarchy of concepts. A general life prediction method for components under creep, fatigue and creep-fatigue conditions is proposed. Fatigue, creep and creep-fatigue data of a typical austenitic stainless steel (i.e., 316) are integrated. Conventional machine learning models (e.g., support vector machine, random forest, Gaussian process regression, shallow neural network) and deep learning model (e.g., deep neural network) are applied for life predictions. Results show that deep learning model exhibits better prediction accuracy and generalization ability than conventional machine learning model. [ABSTRACT FROM AUTHOR]