CHANGES IN ISO/IEC 17025:2017 – SAMPLING UNCERTAINTY?

« Back to Previous Page
21
0

Am I wrong or ISO/IEC DIS 17025:2016 proposes in clause 7.6.2 that laboratories performing sampling MUST evaluate “sampling uncertainty”? I...

Please to read the entire article.

Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 14, 2017 12:00 am
81 views
0
Private comment
Dear Alexis, I would say NIL NOVI SUB SOLE ! It was already specified under the 2005 version even in the general paragraph 5.1.1 and considered by the accreditation process.
Your point 1 is not clear. Most of the accreditation bodies considered that the sampling labs are acting as the main operators. They have the responsibility of the whole measurement process. And they can sub-contract the analyses to an accredited lab. So the whole measurement shall be covered by the accreditation. Same rules apply to the sampling step and the analysis. The analytical lab is only in charge of his part. But the main operator shall consider all of the contributions to the global uncertainty, including those relating to the sampling.
Marked as spam
Posted by olivier NOISET (Discussions: 0, Comments: 9)
Replied on March 13, 2017 8:00 pm
0
Private comment
I agree with you that the sampling uncertainty is a poor estimation very often. Combination of TYPE B components through a bottom up approach where you have any few validation data available for your sampling device.
That estimation of the sampling uncertainty is maybe not the major concern in regard to the sample representativity. Knowing that, even in the framework of some legal controls, only one sample is required !
Marked as spam
Posted by olivier NOISET (Discussions: 0, Comments: 9)
Replied on March 13, 2017 8:00 pm
0
Private comment
Dear Olivier:

One thing is, as we all know, that sampling is a serious potential source of uncertainty in a measurement, and another is to say that evaluation of sampling uncertainty is a widespread practice, managed and required by accreditation bodies.

To say that ISO/IEC 17025:2005 already “required” sampling uncertainty to be determined, seems to me “farfetched”, to say the least.

What 5.1.1 states is “Many factors determine the correctness and reliability of the tests performed by a laboratory. These factors include contributions from: - sampling (5.7);” (among others).

One thing is that and another is to interpret this statement as a "requirement" to evaluate sampling uncertainty. On the other hand, ISO/IEC DIS 17025:2016 does indeed require in 7.6.2 to evaluate sampling uncertainty.
(Continues in next post...)
Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 14, 2017 8:00 pm
0
Private comment
(…comes from previous post)
From your comment one can assume you know of many labs which are evaluating sampling uncertainty but, I must confess, I don´t know anyone who does it.

You are right about that everybody who takes accuracy seriously must be concerned and worry about the “quality” of the sampling, trying to get a sample as representative of the bulk as possible.

Another fact is that, from my experience, most accredited labs get their samples from their customers, who most frequently are not “a sampling lab” and much less an accredited one. What do you suggest the analytical lab should do in those cases? That´s what I was talking about in my item Nº 1.

Best Regards,
Alexis Oramas
Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 14, 2017 8:00 pm
0
Private comment
Surly, for a measurement uncertainty to be taken seriously there must be consideration given to ALL potential sources of uncertainty, both type A and B, which will both exist in some form in any measurement. An uncertainty estimate that is devoid of one type or the other is simply incomplete, which then leads us to ask questions about the traceability of the observation.

It might be some sources are too small to have an influence on the overall uncertainty, and that's fine, but that still needs to be documented and argued even if omitting from the final budget estimate.
Marked as spam
Posted by Eugene Organ (Discussions: 0, Comments: 89)
Replied on March 14, 2017 8:00 pm
0
Private comment
Imagine two different first-party labs, in different factories, each of which carries out sampling in the respective production line of its factory and then analyses those samples, being both accredited for the two activities. Now, according to the DIS, they will have to evaluate sampling uncertainty, as well as that for the actual measurement.

Then, one of those two labs has the “clever” idea of making an arrangement with the production manager so, from now on, the production department is to do the sampling and then take the samples to the lab to be analysed. So, this lab can claim it doesn´t carry out sampling, thus it isn´t compelled to evaluate sampling uncertainty and simply continues evaluating measurement uncertainty.

Meanwhile, the other lab, which continues to carry out its sampling, is now compelled to evaluate them both. Do you get the picture? Isn´t this unfair?

“Isn´t this some kind of discrimination?”

Hope you get now my point.

Best regards,
Alexis Oramas
Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 15, 2017 8:00 pm
0
Private comment
'Quality will out'...

I get the concern, but I suspect a company playing those kind of games will end up catching itself out and would be self defeating in the long term... I'd be very concerned about a production department being given those kind of powers... If I was looking for such a company to do work for me, I'd be steering well clear from the second company as it has put the foxes in charge of the chicken run!
Marked as spam
Posted by Eugene Organ (Discussions: 0, Comments: 89)
Replied on March 15, 2017 8:00 pm
0
Private comment
Dear Alexis,

I understand clearly your opinion. The sampling uncertainty is determined for years in some accreditation activities. Maybe you are confronted to another reality in your field of expertise. But I confirm that is is done by all the labs involved in air pollution controls from stationary sources, for example.

Now about the potential discrimination you mentioned, I think that almost all the labs confronted to customer sampling would state in the report smth like 'the result is only relating to the specific sample provided' and would not take any responsibility in terms of bias (representative sample or not) or associated uncertainty. The point is what is the weight of such statement in a report for the legislator or the control authorities ? Are they able to express more confidence to the other lab covering all the measurement process (and reporting results with larger uncertainties) ? We are talking here about the discrepancy between accreditation and legislation
Marked as spam
Posted by olivier NOISET (Discussions: 0, Comments: 9)
Replied on March 15, 2017 8:00 pm
0
Private comment
Sampling is a core part of the analytical testing method and simply the intermediate precision evaluation can give you the information that includes the variability among replicates that are collected and analyzed (repeatability uncertainty)
However the important part is the sample preparation in the lab prior to analysis
if the sampling is wrong everything is a blunder .
in any analytical testing it is the overall uncertainty in your final result when processed through a complete method that includes the uncertainty due to
sampling is important one cannot estimate the uncertainty only due to sampling as whoever has done the sampling the final analysis result is achieved only after processing through all the steps of the method
Marked as spam
Posted by Khalid Mohiuddin (Discussions: 0, Comments: 10)
Replied on March 16, 2017 8:00 pm
0
Private comment
Sampling uncertainty measurement is essential and mandatory if the laboratory want to be accredited in sampling. The estimation of sampling uncertainty is derived from the process used for estimating overall uncertainty. As Nordtest and Eurachem guides propose, sampling uncertainty is calculated from overall uncertainty and analytical uncertainty. A sampling AND testing laboratory needs to calculate the uncertainty for the analytical procedure alone (with CRMs, RMs etc) and then calculate the overall uncertainty using real samples and different approaches (duplicate samples, sub-sampling etc).

I think we should realize the division implied by "or" in paragraph 7.6.2 of ISO/IEC DIS 17025. It states "sampling OR testing". If a lab is performing only testing, analytical uncertainty must be estimated/calculated. If a lab is performing only sampling, they have to send the samples somewhere (to a subcontractor). The subcontractor needs to have their analytical uncertainty calculated.
Marked as spam
Posted by Efstratios Karyofyllidis (Discussions: 0, Comments: 13)
Replied on March 16, 2017 8:00 pm
0
Private comment
Dear fellows:

I never imagined there could be so much to consider and talk about sampling, as it can be concluded from Eugene´s, Olivier´s, Khalid´s and Efstratios´s latest comments.

First of all, I quite agree with Eugene that sampling is a complex and critical activity to be left in hands of the production or process department but, to my experience, this is something encountered much too often, even for first party labs.

Then Oliver is right that analytical labs try to overcome this fact by stating that note he comments about in his last post. As for his other point, relating to how legislator or control authorities could feel about facing a statement of that kind in a report, I think that´s something difficult to solve, unless we see in the future a new ISO/IEC 17025 compelling all labs to do their own sampling or to only process samples taken by "sampling accredited labs", though many surely might think this is a rather extreme solution.

(Continues in next post…)
Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 16, 2017 8:00 pm
0
Private comment
(…Comes from previous post)
On the other hand, what Khalid put forward is a very important fact: in which of the two uncertainties, sampling´s or measurement´s, is included the contribution made by sub-sampling and/or sample treatment? That´s another difficult one, though I am inclined to think that the contribution from sample treatment would be reflected in the measurement´s repeatability. Other considerations welcomed!!!

Finally, Efstratios is right that the text in clause 7.6.2 “A laboratory performing sampling OR testing activities shall evaluate measurement uncertainty” might be understood as a laboratory should do one activity or the other, but not both. I personally don´t understand it that way, but well could be. One solution to this “possible” confusion is to write in the standard “and/or” instead of “or”.

Best regards to you all,
Alexis Oramas
Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 16, 2017 8:00 pm
0
Private comment
Alexis: 17025 applies to Calibration labs, and to labs which must statistically sample from a lot (before they test). Sampling Uncertainty is thus a very important aspect and in some cases it is the largest factor contributing to Overall Uncertainty. However, any Uncertainty estimation is meaningless unless the test result is a numerical one, and it happens to be close enough to the Acceptance Limits specified by the test standards, so that the application of Uncertainty makes a difference in judging if the test sample passed or failed, or if it is suitable for use in a certain circumstance or not.
Marked as spam
Posted by Vijay Ruikar (Discussions: 0, Comments: 2)
Replied on March 17, 2017 8:00 pm
0
Private comment
Alexis: I personally work in a laboratory that performs both sampling and analysis. My comment on the division between sampling and analysis wasn't meant to discretely distinguish between the two, as labs may perform both sampling and analysis. However, numerical determination of sampling uncertainty may be mandatory depending on the matrix/material being subjected to testing. This applies to all kinds of laboratories: external labs, labs functioning inside factories and so on. I totally agree with Vijay's comment, regarding possible non-compliance to regulatory or QC limits resulting from large sampling uncertainty.
Marked as spam
Posted by Efstratios Karyofyllidis (Discussions: 0, Comments: 13)
Replied on March 18, 2017 8:00 pm
0
Private comment
Dear Efstratios:
I get your point. But, as you comment “numerical determination of sampling uncertainty may be mandatory depending on the matrix/material being subjected to testing”, could you please give us a few examples when sampling uncertainty evaluation wouldn´t involve “numerical determination”, as you called it?
Regards,
Alexis Oramas
Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 19, 2017 8:00 pm
0
Private comment
Thanks for pointing this out Alexis Oramas Pérez. I didn't pick it up when I did my review. I agree that it will be interesting to see how accreditation bodies apply this in their assessments and whether technical assessors have the appropriate competency to review what labs are presenting them!
Marked as spam
Posted by Sonja McFarlane (Discussions: 0, Comments: 3)
Replied on March 19, 2017 8:00 pm
0
Private comment
A sample has to be representative nothing else. Otherwise the sample is not worth to be analyzed in the laboratory. This would be wasting money. Variation from sampling, expressed as uncertainty is not to be mixed up with uncertainty from measurement. Check for the definition of the term measurement uncertainty as used in ISO-GUM. How can one introduce uncertainty before any measurement has been done. We do not know the four dimensional distribution function (time, space) of the characteristics of the measurand. We just assume that the sample represents e.g. the content of a substance correctly. Without any assumption we can in the majority of cases not tell anything about how substance is distributed in the object under investigation from which a sample has to be taken for analysis. This is probably a field for fortune-tellers, but not of scientists.
Marked as spam
Posted by Samuel Wunderli (Discussions: 0, Comments: 9)
Replied on March 19, 2017 8:00 pm
0
Private comment
Dear all,
As Efstratios Karyofyllidis has said, in some cases sampling uncertainty could be numerically calculated from statistical laws, but only in those processes or studies that supply "homogeneous" (without spatial dependance, such a soil sample and "stables" (without time dependance, such a air sample)sub-samples. In fact, such uncertainty is usually clearly detailed in some solid CRMs, including the mínimum sample to be taken to reduce sampling uncertainty compared to the certified uncertainty value.
In some other cases, I consider that sampling uncertainty calculation must be far from the aim of the ISO/IEC DIS 17025, as the effort to be done to "evaluate" (rather than calculate) the sampling uncertainty is huge for analytical laboratories and can be easily rebuttable by accreditation bodies.
Marked as spam
Posted by Alberto José Quejido Cabezas (Discussions: 0, Comments: 2)
Replied on March 20, 2017 8:00 pm
0
Private comment
Dear Alberto José:

It´s good to hear that somebody thinks sampling uncertainty evaluation, if done properly, is not a simple and pleasant walk through the Rose Garden!!!

That´s exactly my point in questioning the “urgent” need to discourage people who is trying to convince his boss, his lab´s fellows, and even himself, that accreditation is worth the effort. Couldn´t ISO/CASCO wait till next edition of the standard to fit in this requirement?

This is particularly so when in relation to measurement uncertainty there is still so much to do. For example, many CRMs suppliers in the US still refuse or avoid to declare the uncertainty of the values “certified” to their RMs. They still insist that with a statement like “this RM ―or the calibration of any standard or instrument, for that matter― is traceable to NIST, without not even mentioning to which NIST standard o CRM in particular they are talking about, they can get away, without a real demonstration of traceability.

Best regards.
Marked as spam
Posted by Alexis Oramas Pérez (Discussions: 1, Comments: 18)
Replied on March 21, 2017 8:00 pm
0
Private comment
Can any one send me the drafted copy of the standard in my mail id: bp_mohanty2000@yahoo.co.in
Marked as spam
Posted by BIBHU PRASAD Mohanty (Discussions: 0, Comments: 3)
Replied on March 21, 2017 8:00 pm
0
Private comment
I do not think that a sampling and testing laboratory could present a trusted pocedure for estimating the sampling uncertainty. Sampling is of a great importance. I do not deny that. But still, algorithms which come from the probability theory could give a result which is even bigger than the measurment result. Almost every test protocol has a note that says that the obtained results refer only to the test sample. I think that sampling uncertainty should be taken into account only in the cases when taken sample is the whole required test quantity and when it has been done with calibrated technical equipment.
Marked as spam
Posted by Valentin Belovski (Discussions: 1, Comments: 4)
Replied on March 21, 2017 8:00 pm
« Back to Previous Page