In the structural integrity assessment of an embrittled reactor pressure vessel (RPV) based on probabilistic fracture mechanics (PFM), it is crucial to consider both of the mean value and the uncertainty of the embrittlement prediction. Typically, the embrittlement prediction uncertainty is given by a normal distribution, and its standard deviation is determined from the residuals of the measured and predicted values of all data used to develop the embrittlement prediction method. Therefore, the same uncertainty is assumed regardless of neutron fluence and the available data. To overcome this issue, the Japan Atomic Energy Agency recently developed the embrittlement prediction method based on machine learning and Bayesian statistics, i.e., the Bayesian nonparametric (BNP) method, and introduced it to the PFM analysis code. The BNP method can predict the probability distribution according to data scarcity and provide more reasonable uncertainty because it estimates significant uncertainties where the scatter of actual measured data is large and the number of data is small. In this study, PFM analysis for a Japanese model RPV in a pressurized water reactor is conducted using the PFM analysis of structural components in aging light water reactors, where the embrittlement probability distribution calculated using the BNP method is used. Furthermore, the effects of different embrittlement prediction methods and uncertainties on the failure frequency of RPVs are investigated in the PFM analyses, and the results are presented.
Read full abstract