Analysis Approaches for Estimating Probability of Detection from Multiple Data Sources (306556)*Aimee Elizabeth Holmes, Pacific Northwest National Laboratory
Bruce P Lin, United States Nuclear Regulatory Commission (NRC)
Ryan M Meyer, Pacific Northwest National Laboratory
Keywords: probability of detection, POD, NDE
This talk will focus on analysis techniques for deriving the probability of detection (POD) for qualified ultrasonic examination in U.S. nuclear power plants (NPP). POD is derived from empirical nondestructive examination (NDE) studies performed in laboratory-like environments on test blocks that simulate relevant NPP component geometries. Data from the U.S. nuclear industry’s Performance Demonstration Initiative (PDI) and two international round robin studies, Program for Inspection of Nickel Alloy Components (PINC) and the Program to Assess the Reliability of Emerging Nondestructive Techniques (PARENT), are examined.
Several approaches to fitting reasonable POD (using logistic regression) when sample sizes are constrained by limited test specimens will be highlighted. How these limitations impact the POD confidence bounds will be shown. A discussion of analysis results, where they agree and differ, and the likely sources of those differences will also be presented. Effective approaches to accommodate combining data from different sources will be described.