STATISTICAL SIMULATION OF FRACTURE TOUGHNESS IN SEGREGATED RPV STEEL USING DEEP-LEARNING-BASED RANDOM FIELD GENERATION AND HIGH-FIDELITY FEA MODELING
Regis KenkoGrand Ballroom C
Charpy impact tests are used in the nuclear industry to certify forging processes. However, the results of these tests may exhibit a strong variability in the context of large metal parts manufactured by Framatome. Preliminary studies have shown that the steel is highly heterogeneous at the millimeter scale in certain areas of forged parts. These heterogeneities are surmised to be the main cause of the variability observed in the results of impact tests. The aim of this study is to qualify and numerically quantify the effect of these heterogeneities on the distribution of fracture energies thanks to an innovative computational approach featuring deep learning to generate 3D realizations of the mechanical properties from sparse experimental results, and high-fidelity modeling of brittle fracture in heterogeneous Charpy specimens.