Application of Failure Event Data to Benchmark Probabilistic Fracture Mechanics Computer Codes

This paper describes an application of data on cracking, leak and rupture events from nuclear power plant operating experience to estimate failure frequencies for piping components that had been previously evaluated using the PROLOCA and PRAISE probabilistic fracture mechanics (PFM) computer codes. The calculations had addressed the failure mechanisms of stress corrosion cracking, intergranular stress corrosion cracking and fatigue for materials and operating conditions that were known to have failed components. The first objective was to benchmark the calculations against field experience. A second objective was a review of uncertainties in the treatments of the data from observed failures and in the structural mechanics models. The database PIPExp-2006 was applied to estimate failure frequencies. Because the number of reported failure events was small, there were also statistical uncertainties in the estimates of frequencies. Comparisons of predicted and observed failure frequencies showed that PFM codes correctly predicted relatively high failure probabilities for components that had experienced field failures. However, the predicted frequencies tended to be significantly greater than those estimated from plant operating experience. A review of the PFM models and inputs to the models showed that uncertainties in the calculations were sufficiently large to explain the differences between the predicted and observed failure frequencies.