ABSTRACT
Nuclear grade 316 stainless steel (SS) is the main material of core internals for liquid sodium-cooled fast reactor (SFR). However, very limited mechanical properties of nuclear grade 316SS under elevated-temperature liquid sodium, including yield strength (YS), ultimate tensile strength (UTS) and total elongation (TE), can be obtained due to long-time consumption and extreme testing condition. Therefore, it is necessary to use limited experimental data to predict mechanical properties of nuclear grade 316SS reliably and efficiently for very long life design up to 60 years. The standardized Euclidean distance was introduced to the radial basis function neural network (RBF-NN) model to develop an improved RBF neural network (IRBF-NN) model, which was trained to solve the problems of back propagation neural network (BP-NN) model. Additionally, the validity of (YS, UTS, TE) about the IRBF-NN model and BP-NN model is evaluated and compared by the absolute relative error (ARE), T-test, F-test, correlation coefficients (R), average absolute error (MAE) and standard deviation (σ). Results clearly illustrate that the artificial neural network (ANN) model is suitable for predicting the mechanical properties of nuclear grade 316SS under elevated-temperature liquid sodium, and the prediction effect of the IRBF-NN model is better than that of the BP-NN model.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (51975424).
Disclosure statement
No potential conflict of interest was reported by the author(s).