Estimating probability of detection from in-service inspection data

Abstract

Due to the complexity of real structures and in-service environment, the probability of detection (POD) curve generated from a laboratory environment and simple coupon samples may not be representative of in-service nondestructive inspection capability and experience. A study was carried out to assess the capability of a recent Berens model to estimate the POD using in-service inspection data. This paper presents the Berens model as well as the results of five case studies, where two types of mean POD curves were estimated and compared with each other. The first POD estimation used the standard POD method, described in the Military Handbook 1823, and used both hit and miss (detected and nondetected cracks) data. The second POD estimation used the Berens model and only hit (detected cracks) and percentage of crack detection (number of detected cracks per number of inspected sites) data. It is shown that the a90 values estimated by the Berens model are close to those from the first POD approach, especially when the percentage of crack detection is known. The study demonstrated the possibility of using the Berens model for estimating an “effective” POD curve from in-service data, where often not all data required by the standard POD approach are available.