Models for non-destructive inspection data

This paper describes the use of statistical models for the evaluation of the reliability of non-destructive inspections. Probability models for the uncertainty in the flaw size determination and the flaw detection are constructed. The flaw sizing models are based on the assumption that the measured and true flaw sizes are related in a simple way: two models based on logarithmic and logit transformations of the flaw sizes are considered. The probability of flaw detection is assumed to depend on the flaw size either through logarithmic or logit transformation. To illustrate the use of these models, they have been applied to flaw sizing data and flaw detection frequency data. Furthermore, models for Bayesian updating of flaw size distributions were developed. In this connection, models for updating the size distribution of a single flaw as well as for handling multiple flaws are introduced.