Abstract
A new methodology is developed for the prediction of material behavior, such as aging processes, by utilizing a combination of domain models and non-linear estimators including neural networks and nearest neighbor regressions. This methodology is applied to the problem of predicting embrittilement levels in light-water reactors by combining the existing models with the conventional non-linear estimators. The Power Reactor Embrittlement Database is used in this study. The results indicate that the combined embrittlement predictor achieved about 56.5% and 32.8% reductions in the uncertainties for General Electric Boiling Water Reactor plate and weld data compared to Regulatory Guide 1.99, Revision 2, respectively. The implications of irradiation temperature effect to the development of radiation embrittlement model are then discussed.
