08:15 Oct-26-2002 Terry Oldberg Engineering, Mechanical Electrical Nuclear Software Consultant, USA, Joined Oct 1999 ^{42}

Protocol for Measurement of POD?

A member of the nde.net forum wrote to me by email:

Interesting to read your comments on POD over the years on the net. Do you have such a thing as an IDIOTS guide (PROTOCOL), on how you feel is the best way to conduct a POD exercise.

I responsed:

Thanks for the interesting question.

There is a situation in which a protocol used frequently in science is available. The situation is that a partition of the material that is inspected by a test, "Partition A," is defined and is fixed as a property of this test. The purpose of the test is to measure the value of one or more properties of each element of this partition. A test is referenced that establishes the true value of each measured property, "Ground Truth".

Partition A forms a legitimate, statistical population for purposes of measuring the reliability of the test. Now, define a partition of the values of one of the measured properties into two parts, "partition B". Define one of these parts as "positive" and the other as "negative". All of the permutations of measured and true values can be now represented by events conventionally termed "true negative," "false negative," "true positive," "false positive," Estimate the false positive and false negative error probabilities by post inspection random sampling in the population, followed by establishment of the ground truth in the sample.

This protocol is inconsistent with your desire to estimate the POD. What you might be able to do is adopt this protocol and use it to measure a property that contains nearly the same information as the POD.

For example, if it is possible to place the boundaries between the elements of Partition A in places where it is extremely unlikely for a flaw to cross a boundary, a probability that captures similar information can be estimated for a test that measures the size of the largest flaw in each element of Partition A; sizes above "z" reside in one of the two elements of Partition B and the remaining sizes reside in the other element. With "positive" defined as values at or above z and z set at a "relatively low value," the probability of a true positive contains information that would be almost identical to the information in the POD, for many purposes; by "relatively low value," I mean a value that is a bit below the value at which the largest flaw becomes capable of propagating to failure.