An overview of the principles of modeling charpy impact energy data using statistical analyses

Abstract

Integrity assessments of Magnox nuclear reactors with steel pressure vessels quantify the temperature margins between the operating temperature of the plant, at any given location, and the onset of upper-shelf temperature. The onset of upper-shelf temperature can be estimated from the fracture toughness properties of each material used in the construction of the pressure vessels. Although start-of-life fracture toughness properties of the materials have been measured, such properties are not available for the neutron-irradiated and thermally aged condition. One of the main effects of neutron irradiation and temperature experienced during service is to increase the ductile-to-brittle transition temperature (DBTT), which can be represented in terms of temperature shifts. In the irradiation surveillance schemes for the Magnox reactors, these temperature shifts can be inferred from Charpy impact energy data which have been measured regularly during the service life. Since Charpy impact energy data are inherently scattered, it is necessary to optimize the interpretation of the data by statistical processing. A recent analysis undertaken by Moskovic et al. concluded that Bayesian analyses are best suited to address the problem. In this overview, we consider the requirements of such analyses and the various options available. We then consider the method proposed by Moskovic et al. with respect to the requirements of the inputs to the integrity assessment and the validity of this approach. In this method of analysis, the distribution of all possible values of model coefficients is established by judging the various possible combinations of these model coefficients in relation to the likelihood of the observed data. Analysis of artificially generated data has been used to compare the effectiveness of Bayesian analyses with those used traditionally.