Quantitative Prediction of Long-Term Precipitation Hardening in Type 316 Stainless Steel Welds Using a Hybrid FEM-Neural Network System

Abstract The stress corrosion cracking (SCC) susceptibility of the weld heat-affected zone (HAZ) in austenitic stainless steels used in nuclear power plants is influenced by several factors such as residual tensile stress, hardness of materials, and environment. This study focuses on low-temperature precipitation hardening (LTPH) due to M 23 C 6 carbide formation in the multi-pass weld HAZ of SUS316 steel during long-term operation. A hybrid prediction system was developed by integrating systematic isothermal aging experiments (300 °C to 500 °C, up to 2000 hours), parametric analysis using the Larson-Miller parameter (LMP), and a neural network (NN) model. The NN model, trained on data across a range of pre-strains (5 to 30 pct), successfully captures the non-linear effects of local strain, thermal history, and nucleation condition on LTPH amount (ΔHV). By coupling this NN model with finite element method (FEM) simulations of multi-pass welding, the spatial distribution of LTPH across the entire HAZ after long-term aging can be predicted. This hybrid FEM-NN approach enables quantitative prediction of the spatial distribution of long-term intragranular hardening in the HAZ of austenitic stainless steels, thereby providing a valuable tool for integrity assessment during long-term plant operation.