A generalized method for the reliability assessment of safety–critical inspection

Reliability assessment of non-destructive evaluation (NDE) inspections is an essential part in developing new inspection techniques where probability of detection (PoD) curves are commonly used as a central reliability metric. The accepted methodology in industry and academia is to use the â-versus-a method and derive the PoD curves analytically based on a set of simplifying assumptions. This study develops a generalized framework that relaxes these assumptions. It is particularly suited to reliability assessments that use numerically simulated data where the number of data points is typically much higher than for assessments based on experimental trials. The generalized framework chooses a regression model for the inspection response from a pool of candidates, estimates the potentially non-constant variance of the response using an average kernel smoother and infers the underlying statistical distribution of the response by a statistical inference algorithm. PoD curves are then calculated analytically and their confidence bounds are obtained by bootstrapping. Numerical results using limited simulated data of an example ultrasonic testing inspection agree well with a reference analysis performed using Monte Carlo integration over a much larger dataset. The developed methodology is a user-friendly framework for reliability assessment of NDE inspections offering enhanced flexibility and robustness.