The Largest Open Access Portal of Nondestructive Testing (NDT)

11:27 Feb-04-2010
Bevel incidence angle (ME paper) and POD

I read with great interest the paper "Bevel Incidence angle requirements for encoded phased arrays" authored by Moles, Kruzic and Ginzel in the last edition of Material Evaluation. Knowing that Michael and Ed are active on this forum, I have a few questions/thoughts I'd like to share.

We happen to work on a POD study for a different application involving the detection of surface or near surface cracks using angled beam UT (fixed angle , no PA). We produced experimental results that are strickingly similar to the results shown in Figure 9 of Moles and al's paper: No clear relationship between amplitude and crack depth (or area) and a large scatter of the values were observed. We came up with the same conclusion regarding the major contribution of the crack angle with respect to the UT beam.

In our case, the "ahat versus a" approach for determining the POD is failing as the method assumes increasing amplitude for increasing crack depth (or area). The question is then how to quantify/demonstrate the reliability of the UT procedure? We are going now for the 29/29 approach used for example by Airbus (see "Application of POD Analysis at Airbus" in this issue of ndt.net). We should be able to prove that we have a 90% POD with 95% reliability for the crack depth defining the acceptance level. However, I feel that this is not totally satisfying: if a technique is very sensitive to a type of defects but shows in average similar response for small and large defects, can it be defined as reliable and can its reliability be quantified?

I would be curious to read if anybody had to work around similar argumentation.

Christophe

16:23 Feb-04-2010
Re: Bevel incidence angle (ME paper) and POD

Christophe
It is interesting to learn that this problem exists in other venues. You indicated that
"ahat versus a" is failing. But as I understand it, this has its underlying principles based on a linear relationship between "ahat" and "a" (sized vs true size). But that is not the same as the amplitude responses having a linear relationship to flaw size. I would think that in order to provide "ahat" and "a" analysis you would need to have some technique for sizing the indications and comparing this to the destructively determined sizes. There is still a threshold to concern yourself with (i.e. some amplitude below which responses are not sized). But if no real sizing has been done and you are simply using the threshold amplitude to indicate detections, then I would think that the Hit/Miss approach is perhaps more appropriate. The concept of Probability of Rejection introduced in DNV OS F101- 2007 seems to fit better with the "ahat vesrus a" concepts.

18:55 Feb-04-2010
Re: Bevel incidence angle (ME paper) and POD

Ed,

You are right: the procedure we are looking at (detection of fatigue cracks in train axles) does not call for sizing and the hit-miss technique is certainly more appropriate. My understanding of the 29/29 method is that it is a corollary of the hit-miss that insures the statistical reliability of the final result for a specific crack size.

We choose the "a hat versus a" for economical reason as the number of sample needed is much lower and also because the end-user is interested in crack size even if sizing is not required . By failing, I meant that the linear regression of the ln(a hat) vs ln(a) gives in our case a zero or negative slope, leading to a failure of the model when computing the POD.

Reformulating my original question, I did not really find any guidelines for choosing the right technique in Berens MIL Handbook on NDE reliability. Your paper and our results show that in some cases, the "a hat versus a" is not valid and I wonder which quantitative criteria should be used in order to assess the best technique for POD evaluation.

Christophe