I appreciate if someone answers my questions.
1-Why a good estimation of a90/95 is so essential for NDT INSPECTION?
2-what is the effect of defect size on the Confidence interval
? and why confidence interval is biggest on hit and miss methods compare to a hat versu a.
3-How many points, I need to consider for noise study on the material? (35 or 38 or more than) because in the "developing qualification NDT procedure book” chapter 11 page 100, mentioned +35 point is needed to build an accurate noise study, but I did not find any standard or procedure about this issue? (why 35 or 38 points? which one is better ?)
4-How can I determine or estimate noise in a material with Civa simulation?
5-For noise estimation, I have to determine the noise level by PAUT on martensite or not?do you have some procedure for it? Can I use UT for it? (which one is correct, easy and accurate- because I want to inspect my weld joints with PAUT).
NDT Inspector, University Professor, Consultant, NDT Inspector, R&D
Department of Mechanical Engineering - Politecnico di Milano, Italy, Joined Apr 2013, 2
Re: POD In Reply to Amir at 07:12 Apr-13-2019 (Opening).
1) for structural integrity, the important thing is the largest defect you can miss, not the smallest you can detect
2) confidence depends on the standard deviation and the number of responses, not their mean value. "Hit/Miss" is less informative than "signal response", so it is more uncertain
3) the question is not clear
4) you can't. In CIVA, noise is an input, not an output
5) the question is not clear
R & D, -
Materials Research Institute, Canada, Joined Nov 1998, 1300
Re: POD In Reply to Amir at 07:12 Apr-13-2019 (Opening).
This is a timely question. I have just returned from an excellent course on NDT Reliability given by Extende (distributors of Civa). I suspect that many of your concerns could be addressed by attending this course.
I will attempt to comment on each of your items.
1. The origin of the 90|95 used in NDT seem to be related to the foundational work on the topic by NASA (in the USA). The report “The Detection of Fatigue Cracks by Non-destructive Testing Methods, W.D. Rummel, et al, CR-2369, 1974”. In that paper 2 groups of samples were identified. The larger (60 observations) was capable of using a 95% reliability at 95% confidence and the smaller (30 observations) corresponded to a 90% reliability at 95% confidence based on no failure. This came directly from the binomial treatment of the data used. It is worth noting that Mr. Rummel has contributed many papers to the NDT community including on NDT.net (e.g. https://www.ndt.net/article/wcndt2012/papers/608_wcndtfinal00607.pdf). Another paper he contributed to (https://www.ndt.net/article/wcndt00/papers/idn669/idn669.htm) illustrates how noise is treated.
2. You asked “why confidence interval is biggest on hit and miss methods compare to ahat versus a.” I suspect you are asking about the flaw size assessed at the 50% CI versus the 95% CI. I am not sure that you can make such a generalised comment. Not all NDT data can be used to generate an ahat vs a POD. And even when you can find some relationship between the response amplitude and flaw size, there are multiple techniques used to calculate confidence intervals just as there are multiple techniques to derive the POD curve fitting models.
3. The number of points for a noise study seems to be a trivial concern. It should be very easy to collect samples for a noise assessment. You quoted Michael Wright’s book “Developing Qualified NDT Procedures” where he stated that “a minimum of 35 points is recommended”. I am not sure where he obtained this recommendation, but he does go on to qualify the statement with this number being due mainly to the normal approximation accuracy at 35 points. Michael makes a personal observation that obtaining noise data is cheap and suggests that 50+ noise samples be used.
4. Civa 2017 provides an analysis capability when structural noise is applied. You can use Histogram Tool to initialize an area and a level of noise as a reference. To create a Noise level reference, apply the “SNR measure” button. A new SNR panel is created in the “Palette/Gain” toolbox, and the current colormap is adapted. The noise level is displayed in the CIVA manager data tree, and all amplitude measurements are given in reference to the noise level (0dB = noise level).
5. The level of noise that you would model in Civa is easily obtained from the actual samples you plan to work with. I suppose you could use monoelement probes to obtain a noise level relative to your reference sensitivity. But if you plan on using phased-array UT why would you bother with the monoelement noise acquisition? In an ultrasonic test, noise is usually a result of backscatter from grain structure and to some extent the electronic jitter mostly attributable to the receiver amplifier. If you plan on using a sectorial scan on your martensitic weld you could easily capture S-scans from several places along unflawed sections of the weld. The process of angulation of the beam usually results in less pressure being transferred at higher refracted angles. As a result, you typically add more receiver gain to the delay laws at higher angles so electrical noise at higher angles is likely to be greater than at the lower angles (as well, at higher angles you may start to see noise associated with the couplant at the wedge/metal interface). Another reason to use the actual phased array probe rather than one or two monoelement probes of the same aperture is that the circuitry of a phased-array instrument is different from a single channel monoelement instrument. Remember also that instruments have receiver filters which can have an effect on the observed noise.
Having collected S-scans for the noise assessment in your material you may want to observe the noise level of the A-scans at each angle because it may vary. You may then need to apply different thresholds of noise for each angle. If the noise level is well below the evaluation threshold that you require for your flaw detection, the concerns for noise may be unwarranted.