Directed Design of Experiments for Validating Probability of Detection Capability of Nde Systems (Doepod)

Abstract

The capability of an inspection system is established by applications of various methodologies to determine the probability of detection (POD). One accepted metric of an adequate inspection system is that there is 95% confidence that the POD is greater than 90% (90/95 POD). Directed design of experiments for probability of detection (DOEPOD) has been developed to provide an efficient and accurate methodology that yields observed POD and confidence bounds for both Hit-Miss or signal amplitude testing. Specifically, DOEPOD demands utilization of observance of occurrences. Directed DOEPOD does not assume prescribed POD logarithmic or similar functions with assumed adequacy over a wide range of flaw sizes and inspection system technologies, so that multi-parameter curve fitting or model optimization approaches to generate a POD curve are not required.

Statistical Analysis of Hit/Miss Data (Preprint)

“…These two developments were incorporated in the second edition of MIL-HDBK-1823. Recently, concerns have been raised about the current approach, and binomial methods have once again been proposed to mitigate concerns about the behavior of POD for flaw sizes greater than the established a 90 estimates generated by parametric approaches (Generazio, 2011). Further work on nonparametric methods has been done by Spencer with the assumption that the POD is a continuous non-decreasing function of flaw size (Spencer, 2011).…”

“…In the authors' opinion, there is a valid concern raised in ( Generazio, 2011), and that is that one could analyze hit/miss data that doesn't meet the requirements of MIL-HDBK-1823A and still generate an estimate for a 90/95 that could be misused. Another concern in (Generazio, 2011) is the assumption of a monontonically increasing POD function.…”

“…Before applying 3 and 4 parameter models to difficult data sets in (Generazio, 2011), Bayesian methods are illustrated on data from Berens seminal paper on the topic (Berens, 1989). For hit/miss analysis the data is assumed to follow a Bernoulli distribution.…”

“…If the 3 or 4 parameter model is selected, it is very possible that a 90 or a 90/95 might not exist even though inference with the 2-parameter model provides estimate. Let's look at an example that appeared in (Generazio, 2011). There was a particular set of data that the author identified as a "Case 2 data set D8001(3)L." It was reported that the "Logit-ML" a 90/95 value was 12.95 mm.…”

Detection, 3-D positioning, and sizing of small pore defects using digital radiography and tracking

“…Others [27] try to approximate the confidence intervals with the use of sophisticated DOE methods combined with less strict assumptions on the NDE system. However, in this article, the POD and the 95% lower confidence interval (POD95) methods of [28] is used.…”