Using requirements on merit ratios for assessing reliability of NDE flaw detection in multi-hit detection in digital radiography

Abstract

Current approaches to qualification or reliability assessment of NDE procedures by traditional probability of detection (POD) testing per MIL-HDBK-1823A require significant number of specimens with known size real or simulated flaws. For many practical applications in aerospace industry, either it is difficult to produce or it is economically not feasible to fabricate the necessary flaw specimens in required quantity. Thus, there is a strong need for alternative methodologies to provide assessments of reliability of NDE procedures using fewer flaw specimens. The paper provides an alternate NDE reliability assessment approach using dependence of the probability of detection (POD) and probability of false positive (POF), on the contrast-to-noise ratio (CNR) and decision threshold-to-noise ratio (TNR) for selected POD and POF models. The NDE reliability assessment approach is termed as limited sample (LS) POD assessment. LS POD approach for single hit flaw detection is developed first. Current work provides some progress in extending LS POD single hit approach to LS POD multi-hit approach. In multi-hit flaw detection, flaw detection data is mapped to form an indication with cluster of pixels in a 2D image which is absent in single-hit flaw detection. Resolution of the measurement system is additional factor to be accounted for in multi-hit LS POD. Resolution is addressed with a term defined as resolution ratio. These limits can be determined through modeling but should also be corroborated through empirical data. In the LS POD approach, the signal response from a fixed size flaw of interest is assumed to follow a normal distribution. The approach uses statistical tolerances computed using signal response sample data and 𝑘₁-factor. Similar to traditional POD analysis, goal of the multi-hit LS-POD approach is to determine POD, except emphasis is on relationship of resolution ratio and CNR with POD and POF. The flaw size is determined from the resolution ratio. The POF is also estimated in the analysis. Noise is assumed to follow either a normal or lognormal distribution. For reliable detection of the target size flaw, minimum POD/Conf. of 90/95% and maximum POF of 1% are assumed. The sample of signal responses is assumed to be representative of the assumed population of signal responses. Similarly, the noise measurements are assumed to be representative of those expected in the inspection of real hardware. Using simulation, current work indicates that resolution ratios relate to minimum CNR for reliable flaw detection. An idealized case of a round x-ray indication is chosen for the simulation. Empirical approach for multi-hit POD analysis is provided.