A physics-informed neural network for creep life prediction of austenitic stainless steels in air and liquid sodium

Creep life prediction of component materials exposed to air and liquid sodium environments is critical to ensure the safe operation and structural integrity of a sodium-cooled fast reactor. In this paper, a method for predicting the creep life of a wide range of austenitic stainless steels in air and liquid sodium was proposed based on a physics-informed neural network. Based on the established datasets for sodium corrosion rates and creep life in air and liquid sodium, the predictive performance of physical equations, conventional machine learning models, and the proposed model were assessed. Subsequently, a data-driven creep life assessment framework was established, providing insight into the engineering application of machine learning methods in high-temperature structure assessment. The results show that the creep fracture of austenitic stainless steel is accelerated by liquid sodium corrosion. The proposed physics-informed neural network exhibits enhanced suitability and accuracy for predicting the sodium corrosion rate and creep life than physical equations and conventional machine learning methods. A data-driven creep life prediction method for metals under corrosion was proposed. New physical constraint and feature have been proposed based on domain knowledge. The excellent predictive performance of proposed method has been validated. Data-driven creep assessment under environment effects has been proposed.