Surrogate model development and validation for reliability analysis of reactor pressure vessels

In nuclear light water reactors, the reactor core is contained within a thick walled steel reactor pressure vessel (RPV). Over time, material embrittlement caused by exposure to neutron flux makes the RPV increasingly susceptible to fracture under transient conditions. Because of parameter uncertainties, probabilistic methods are widely used in assessing RPV integrity. For efficient probabilistic analysis, techniques to rapidly evaluate the stress intensity factor for given flaw geometry and stress conditions are essential. The stress intensity factor influence coefficient (SIFIC) technique is widely used for this purpose, but is limited to axis-aligned flaw geometries. To consider a wider range of flaw geometries, surrogate models to compute stress intensity factors are explored. Four surrogate modeling techniques are applied here to compute SIFICs from a set of training data, including two different response surface polynomials, a model utilizing ordinary kriging and another using interpolation. Errors in the SIFICs are assessed for all of these techniques. These techniques are benchmarked against a benchmark solution by computing the time history of the stress intensity factor for an axis-aligned, semi-elliptical surface breaking flaw in an RPV subjected to a transient loading history. All of these techniques compare well with the benchmark solution.