Talk:Nondestructive testing
From Wikipedia, the free encyclopedia
Contents |
[edit] Irrelevant websites
I removed the link to NDT.org.ua - Information Portal of NDT and TD. The website is entirely in Russian. I am also not sure what TD means. User SmArt, please explain the relevance. Perhaps the link would be more useful on Russian Wikipedia.
DavidMack 14:18, 11 October 2006 (UTC)
82.163.171.137 keeps inserting links from John Drury and a UK company called Silverwing. These links are already included under the relevant topics (Magnetic flux leakage and Ultrasonic inspection. They do not make sense on this page or under the heading NDT Published Journals, so I removed them.
DavidMack 21:34, 13 October 2006 (UTC)
[edit] Statistics
This is not a statistical debate forum.
To NDT Guru:
Of course, it's not a statistical debate forum. But it is a discussion, so please quit deleting questions and comments. Math (that includes statistics and probability) is the heart of NDT and any science. Do the math and let the banter begin!
--Terrible Swede 01:03, 28 July 2006 (UTC)
[edit] Reliability not defined, expected utility varies with user
November 26, 2005
In the past few days, I've revised the text to address issues that I believe to be quite important. The pre-revision text did not address the fact that probabilities necessary to the computation of a defect detection test's expected utility are unavailable. Also, it did not address the fact that the expected utility of a defect detection test is dependent upon user's utilities, which vary by user. A result of both facts is that it is impossible to state whether individual defect detection tests are useful or not-useful; the prior text implied that they are useful. I've also edited the text of past contributors in an attempt at making them more concise and clearer.
Terry Oldberg http://www.oldberg.biz
03MAR2006
Mr. Oldberg:
Why don't you use the traditional terms of binomial probability distribution: success (S) and failure (F); instead of OK/not_OK?
--Terrible Swede 18:40, 3 March 2006 (UTC)
- Terrible Swede: Thanks for taking the time to discuss my contribution. From a mathematical perspective, any pair of symbols works. However, the semantics are also important. Nondestructive testing is a measuring process. It seems to me that the property of a part being measured is its all rightness and not its successfulness. Very little of the literature of nondestructive testing deals with its reliability and this results in an absence of terminological norms. In writing my portion of the article, I used as much of the terminology of diagnostic testing in medicine as possible. In medicine, however, the property of the human being tested is healthfulness, but healthfulness isn't a property of a part so I used all rightness (OK is synonymous with all right, not OK with not all right). By the way, the field of defect detection testing has not yet gotten around to defining what is measured. Thus, its understandable that in writing about its reliability one has to invent names for the values of the binary variable that is being measured.Terry Oldberg 00:49, 4 March 2006 (UTC)
03MAR2006
Mr. Oldberg:
Also, if you have determined to use the terms "true positive", "true negative", "false positive", and "false negative", which I would tend to use and are good terms, wouldn't you now need to use a multinomial probability distribution instead a binomial prob distribution?
--Terrible Swede 18:40, 3 March 2006 (UTC)
- Terrible Swede: I'm unclear on the pertinence of the issue of the sample distributions. Please clarify your question.Terry Oldberg 20:43, 3 March 2006 (UTC)
[edit] Unnecessary focus in article
A large portion of this article seems to be devoted to the views of Mr. Oldberg. I do not have the background in nondestructive testing (my background is in math) to support or refute his claims, but it seems that having so much text about what appears to be a minority opinion is disingenuous. If I am wrong, and this is in fact a majority opinion within the field than feel free to contradict me. On the other hand, if I am right then perhaps the article should be rewritten to deemphasize this material, or at least point out that it is not widely accepted. For example, have these results been addressed anywhere in the statistics literature?
[edit] Irrelevant debate in article
I am an engineering physicist who has worked in the NDT field for 20 years, and I do not see the relevance of Mr. Oldberg's contribution. It seems to me that the primary purpose of this article and linked articles should be to explain the various methods of nondestructive testing (NDT). Probability of detection (POD) is an important area of study in NDT, but Mr. Oldberg has concentrated on one very esoteric aspect of POD and has not made an effort to clarify the issue or where his ideas fit in to the big scheme of things.
Davidmack 19:32, 25 July 2006 (UTC)
---
Actually, having read the paper I now have a question for Mr. Oldberg. In your paper, the following two sentences appear:
1) "If a "defect" had an indication within 3 inches of it, he assigned it to a true positive. Otherwise, he assigned it to a false negative."
OK, so we take all the defects, and call them either true positives or false negatives.
2) "However, Bradley's rule of evidence produces a conflict with this assumption. It assigns every "defect" to a true positive if it is within 3 inches of an indication. The same rule assigns every such "defect" to a false negative also. Let us examine the consequences of this ambiguity."
This appears to contradict the previous statement. It also forms the crux of the following argument. How is it that the above way of counting defects assigns defects to two categories simultaneously?
In my understanding of the above procedure, the relative frequency of defects is just 1 (as all trials are counted as defects of one sort or another). The probability of detection is then estimated as n_tp/(n_tp + n_fn), which makes sense to me.
Can you please explain statement (2) further?
(Note: I am not saying that what you've said is incorrect--I legitimately would like to know what you mean. I certainly agree with you that some of the measurements sound dubious from the way you have described them.)
---
Dear anonymous reader
Thank you for giving me the opportunity to comment.
The necessity for assigning a defect to two mutually exclusive categories rather than one seems to be what you don't understand. Assigning an object to two mutually exclusive categories may sound wrong because this violates the principle of Aristotelian logic that is called the "excluded middle." Violations of the excluded middle are built into defect detection tests, including the one reviewed in my 1995 paper.
Imagine an inspection technology in which the inspector provides the X, Y and Z coordinate of a point inside each, purported defect. A rule is in place in this technology which states that a defect is detected if one or more of these points lie inside it. In this case, each defect is either detected or it is not detected. The excluded middle is preserved and a probability of detection is defined. However, I'm not aware of a single defect detection test that operates in this manner. In particular, the test of pressurized water reactor steam generator tubes that is reviewed in my 1995 paper doesn't operate in this manner.
Now imagine that the point is expanded into a spherical region. As it expands in radius, it comes to intersect sound material plus many defects. Neither a single defect nor a single region of sound material is being detected. Rather, they are all being detected simultaneously. In this way, the law of the excluded middle is violated. Probability theory, which assumes the excluded middle, is also violated. This example captures the portion of the content of my paper which is under discussion.
Your proposal for assigning the ratio n_tp/(n_tp + n_fn) to a "probability of detection" fails because your "probability" is not a probability. Probability theory assumes the excluded middle but the test violates the excluded middle by its design.
The "pseudoprobability" with which I replaced probability in reworking the statistics produced by the study of Bradley et al copes with this situation by mimicking the one-to-many relation from detections to detected defects by assigning individual defects to the two categories of false positives and true negatives. A simple way of putting this is that one knows that a defect has been detected but doesn't know which. Pseudoprobability theory is an inadequate response to a incompetently designed technology. The proper response would be for the tests to be redesigned to preserve the excluded middle. The folks who control the content of nuclear reactor inspection technology have proved stoutly resistent to this course for a period of more than 20 years. That a misuse of statistical ideas and terminology is ubiquitous in the field of defect detection testing may be associated with this phenomenon. Competent design has been replaced by fallacious statistical reasoning. --Terry Oldberg 17:16, 19 July 2006 (UTC)
[edit] Removal of Text
I have mainly removed the section on statistics because it does not represent a neutral point of view: it engages in debate and argument, contrary to Wiki policy that "Debates are described, represented, and characterized, but not engaged in." The contributor of that section references his own papers. One subsection begins "Oldberg and Christensen (1995) and Oldberg (2005) report that ...", whereas "report" is listed as a word to avoid because it "bestows a sense of impartiality on a source." Perhaps the original contributor could submit a revised, more neutral version.
DavidMack 23:19, 12 January 2007 (UTC)
[edit] Garbling of message on statistics
In the year and a half since I last participated, the message on the anomalous statistics of NDT has become garbled beyond recognition. The message should be that defect detection tests fail to define statistical populations, thus failing to provide the basis for establishing their reliability. This message is spelled out in several peer reviewed articles, none of them challenged in the peer reviewed literature. If I hear no objections, I'll ungarble the text. If anyone wishes to discuss the technical merits, please contact me at 650-941-0533 or terry_oldberg@yahoo.com. Please be sure to contact me before regarbling the message.
Terry Oldberg, BME, MSE, MSEE, PE Nuclear Engineer —Preceding unsigned comment added by 171.66.222.156 (talk) 00:07, 31 October 2007 (UTC)
[edit] Removing the photo from NDT page
[I have taken the liberty of placing Plenumchamber's message to me on the talk page. DavidMack]
Hi, I see that you have removed my photo that I have taken in my NDT shop indicating that it is a commercial photo. I do not understand your point of view. For example is it a commercial photo if you put an Airbus a/c photo in an article? I have installed the photo in order to give the readers an idea about an test tool. As I have no relations with GE other than being an user, I have no purpose such as doing advertaisement of a company or product. Thanks in advance for your answer back.--Plenumchamber 11:50, 22 June 2007 (UTC)
- Plenumchamber, thanks for explaining your motives, which I understand were not to advertise. And we do need pictures on this article. To me the picture [1] appears commercial because it only shows an instrument with a blank screen and gives the model name. It does not give any hint as to what the instrument does or how it relates to NDT. Do you have any pictures of data or of a technician using an instrument? That way the emphasis is on NDT. We also don't want companies following suit and putting up their instruments. Wiki is quite an advertising magnet. That said, if you want to put the picture back, I won't remove it, and sorry for the inconvenience. — DavidMack 15:13, 22 June 2007 (UTC)
- Thanks for the expl. I will not put back the picture back cause I understand and share your resons, I will try to create a better photo while performing a test action. Thank you.--Plenumchamber 09:13, 25 June 2007 (UTC)
- Thanks for the excellent photos. I hope you approve of the editing I did on the captions. — DavidMack 17:25, 28 June 2007 (UTC)
- Thanks for the edits, now it is better--Plenumchamber 07:43, 2 July 2007 (UTC)
- Thanks for the excellent photos. I hope you approve of the editing I did on the captions. — DavidMack 17:25, 28 June 2007 (UTC)
[edit] Link clean-up
We got caught. Too many links. Apparently Wikipedia is not supposed to be a repository of links. We should link only to 'further reading type material', i.e. "neutral and accurate material that cannot be integrated into the Wikipedia article due to ... amount of detail." So I removed most of our little link farm. I hope everyone is agreeable to this. — DavidMack 19:01, 26 June 2007 (UTC)
Someone added a link to a research institute, so I put back the original list of research NDE institutes. If anyone complains that they are commercial, they'll have to go. — DavidMack 16:51, 24 September 2007 (UTC)
[edit] Microwave based NDT
A lot has happened over the past decade in the domain of microwave based NDT - for applications such as cancer detection (Susan C. Hagness, U. of Wisconsin, et al) and fault detection in concrete. doesn't that merit mention here? 220.225.214.2 (talk) 09:47, 28 February 2008 (UTC)
![[1]](http://en.wikipedia.org/wiki/Image:GE_USM_35X_DAC_LEMO_type_NDT_tool.jpg){kind=link}