A physics-informed neural network approach for predicting fatigue life of SLM 316L stainless steel based on defect features

Abstract

Defects in additively manufactured materials severely limit the performance of parts in practical applications, often exposing them to the risk of fatigue failure. In order to improve the reliability and performance of additively manufactured parts, it becomes crucial to accurately predict the fatigue life of the material. Although traditional semi-empirical formulas can assess the effect of defects on the fatigue performance of parts, they still lack detailed research and consideration of defect morphology features. Therefore, this study proposes a method based on Physics-Informed Neural Networks (PINN). This method improves the predictive capability of the model and enhances its interpretability by extracting the sensitive features of critical defects and embedding known physical knowledge or fracture mechanics methods as loss functions into the training process of the neural network. Additionally, the method effectively captures the complex relationship between defect features and fatigue life, providing a deeper understanding of the model prediction results. The results show that the PINN model considering feature-related knowledge has higher prediction accuracy and reliability, and all predicted fatigue life are narrowed within 2-factor bands, enabling more accurate prediction of fatigue life for SLM 316L stainless steel under different processing conditions.