|NDT.net February 2003, Vol. 8 No.2|
I hope there are some forum members that might be interested in commenting on the issue of sizing tolerances.
Background: Concerns for sizing accuracy are not unique to pipeline AUT. Pressure vessel people have long been concerned for flaw vertical extent and proximity of one flaw to another or proximity to a surface so they can use the interaction rules of fracture mechanics. These concepts were the early foundations of the zone discrimination technique we used in Canadian pipeline construction in the 1980s. CSA Z-662 used the guidelines of BS PD 6493 when developing the fracture mechanics based rules in its Appendix K. Similar rules were developed in API 1104 in its Appendix. But fracture mechanics calculations require tolerances to allow for variation in parameters. One of the important test parameters we deal with in NDT is the accuracy to which we size a flaw. At some point a flaw becomes sufficiently large enough that it is considered a potential point of failure initiation and thereby becomes a "defect".
When fracture mechanics engineers develop their equations, the tolerances are added at each step along the way. Maximum loads (stress), fatigue cycles, wall thickness and metal toughness (as well as other parameters) would all have some variation and these are included in the safety factors of the equations used by the engineers. Some codes (e.g. the ASME Code Case 2235) do not detail the tolerances and assumptions made. They simply provide a table with these tolerances presumably already incorporated. In API 1104 the sizing error is subtracted from the allowable flaw size and the height and length limits derived from that. In PD 6493 they state that for AUT the standard deviation is somewhat better than for manual UT and 2-5mm standard deviation errors were used in the 1991 edition. Where AUT was used (for Level 2 assessment), BS PD 6493 required a 20% partial safety factor be added to the estimated flaw size thereby shifting the curve that defines the allowed length to allow shorter lengths due to sizing uncertainty.
Recently in the pipeline industry there has been pressure to reduce even further the tolerance for sizing error. The simple addition of zones that we used in Canada, which assumes a worst-case scenario, becomes overly conservative when interaction rules are applied so options are needed. API 1104 requires that the sizing technique accuracy has been established, but it does not give guidance on how it is to be done. DNV OS F101 requires a statistical method of tolerance determination. PD 6493 made assumptions based on generic tests.
But the demands on production speed for pipeline construction do not allow for a lot of time to be spent sizing flaws. TOFD and tip diffraction are generally recognized as the best option for accurate size "estimates" but even these techniques are limited by the need to resolve the upper and lower tip signals. This resolution is not possible when the flaw size is on the order of just 1-2 wavelengths because the pulse itself masks the tip responses.
In an effort to maintain the speed of production demanded in pipeline construction, variations on amplitude sizing have been most commonly used in the past 6-7 years. In-house "policies" or sometimes even company "specifications" have detailed how to treat amplitude of signals such that the amplitude can be linearly related to vertical extent.
Concerns for the effects of flaw length and flaw shape on the signal amplitude have been raised by myself and others. But when used as a "policy" the technique of linearising amplitude provides a reasonably conservative treatment of the data while at the same time allowing rapid assessment of the results. Recently several studies have been carried out that will hopefully be published and made available to all concerned. Typically it may be found that, on average, there is an oversizing if about 1-1.5mm (when using 2mm diameter flat bottom holes to set sensitivity as per Canadian Standards). But these same studies also indicate there is also a standard deviation of about 1-1.5mm. These studies provide some proof that the assumptions made in PD 6493 are perhaps a bit conservative at the upper end (5mm standard deviation) but the requirement to add 20% for Level 2 assessments was not too onerous for PD 6493 Level 2 assessments.
New Specifications Imposing Unachievable Tolerances
The above background indicates that the linear scanning we have been using with the zonal discrimination technique has been reasonably conservative and a nominal 1mm error allowance falls very near twice the wavelength used by the AUT systems (0.5-0.7mm). For small weld inclusions on the order of this 1-2 wavelength size even the temporal methods (TOFD and tip diffraction) are not capable of much better accuracies.
Some companies have, as a result of these pressures to "improve" sizing tolerances, made fantastic "claims" to special sizing techniques that they use. But when compared one to the other, there are no "significant" differences in the statistics and we are still left with a standard deviation of 1-1.5mm in the options proposed using amplitude responses.
In the past year the sizing rhetoric has reached what one might consider the ridiculous point. Recently I have read specifications for large corporations based in the Gulf of Mexico and all written by the same AUT expert. These company specifications require that any company providing AUT services "qualify" their system showing that it can meet the requirements of the specification. Such a requirement is normal enough until one reads that the specification requires sizing accuracy to be 0.3mm or better on surface breaking or surface interacting flaws and 0.8mm or better for buried flaws. 0.3mm is on the order of half a wavelength of the pulse being used (5-7MHz)! These tolerances are not even worded as standard deviations! Instead, they are imposed as absolute limits in the specification. Any size estimate made that a macro sectioning shows to be incorrect by an amount greater than the 0.3mm and 0.8mm limits is sufficient to fail a qualification! (Press just 3 pieces of paper together to appreciate what 0.3mm looks like).
In discussions with the parties imposing these requirements I pointed out that not only were these tolerances on the order of the standoff positioning accuracies (0.5mm) and the calibration target tolerances (0.2mm), these tolerances were significantly more stringent than the wall thickness variations. The AUT expert replied that, "Any AUT system that would be approved to the specification would need to be 'Robust' enough to accommodate a 2mm thickness change!"
I have evaluated performance and audited results from all of the AUT systems now commonly used on pipeline construction around the world. None of these systems can "Honestly" make such a claim of accuracy. This would mean that no system in the world could officially "qualify" to such a specification qualification …… unless it was pre-qualified without the need for the same verification requirements as others are imposed with.
The AUT expert that wrote the 0.3mm tolerance for flaw sizing told me that it was proven possible by a scientific paper published by a German author. I requested the source of the paper but it was not provided.
If any others in the forum know of any papers relating to flaw sizing tolerances perhaps they could pass the references on to me. I have collected several papers indicating shortcomings in amplitude sizing accuracy but if someone has published a new method that can provide the 0.3mm tolerance of weld flaws based on amplitude (or any other ultrasonic method) I would be pleased to learn about it.
Until then perhaps others would like to pass on their experiences where the realities of physics do not match the expectations of the specifications.
10 Feb. 2003
|© NDT.net email@example.com|||Top||