In a 2007 paper entitled “Application of Failure Event Data to Benchmark Probabilistic Fracture Mechanics (PFM) Computer Codes” (Simonen, F. A., Gosselin, S. R., Lydell, B. O. Y., Rudland, D. L., & Wikowski, G. M. Proc. ASME PVP Conf., San Antonio, TX, Paper PVP2007-26373), it was reported that the two benchmarked PFM models, PRO-LOCA and PRAISE, predicted significantly higher failure probabilities of cracking than those derived from field data in three PWR and one BWR cases by a factor ranging from 30 to 10,000. To explain the reasons for having such a large discrepancy, the authors listed ten sources of uncertainties: (1) Welding Residual Stresses. (2) Crack Initiation Predictions. (3) Crack Growth Rates. (4) Circumferential Stress Variation. (5) Operating temperatures different from design temperatures. (6) Temperature factor in actual activation energy vs. assumed. (7) Under reporting of field data due to NDE limitations. (8) Uncertainty in modeling initiation, growth, and linking of multiple cracks around the circumference of a weld. (9) Correlation of crack initiation times and growth rates. (10) Insights from NUREG/CR-6674 (2000) fatigue crack growth models using conservative inputs for cyclic strain rates and environmental parameters such as oxygen content. In this paper we design a Python-based plug-in that allows a user to address those ten sources of uncertainties. This approach is based on the statistical theory of design of experiments with a 2-level factorial design, where a small number of runs is enough to estimate the uncertainties in the predictions of PFM models due to some combination of the source uncertainties listed by Simonen et al (PVP2007-26373).

This content is only available via PDF.
You do not currently have access to this content.