A physics-informed neural network for creep-fatigue life prediction of components at elevated temperatures. (December 2021)

• Record Type: Journal Article
• Authors: Zhang, Xiao-Cheng; Gong, Jian-Guo; Xuan, Fu-Zhen
• Is Part Of: Engineering fracture mechanics. Volume 258(2021)
• Publication Date: 2021-12
• Subjects: Machine learning -- Deep neural network -- Physics-informed -- Creep-fatigue -- Life prediction; Fracture mechanics -- Periodicals; Rupture, Mécanique de la -- Périodiques; Fracture mechanics; Periodicals; 620.112605
• DOI: 10.1016/j.engfracmech.2021.108130
• Languages: English
• ISSNs: 0013-7944

Abstract

Highlights: Physics-informed neural network (PINN) is built for creep-fatigue life prediction. Physics-informed feature engineering and physics-informed loss function are applied. PINN exhibits better prediction capacity than conventional machine learning models. Abstract: Physics-informed neural network has strong generalization ability for small dataset, due to the inclusion of underlying physical knowledge. Two strategies are enforced to incorporate physics constraints to a deep neural network in this work. One is to obtain extended features through physics-informed feature engineering, and the other is to incorporate physics-informed loss function into deep neural network as constraints. Conventional machine learning models, deep neural network and physics-informed neural network are applied to predict creep-fatigue life of 316 stainless steel. Results show that physics-informed neural network presents better prediction accuracy than deep neural network and conventional machine learning models.