Measurement Uncertainty & Tolerance - Adding Uncertainty to the Tolerance

Charles Wathen

Involved - Posts
Hi everyone,
I have a question regarding a practice that our Ireland facility has been doing with measurement uncertainty.

For example, they send out an instrument for calibration. It comes back stating that the tolerance was within, but it was right at the edge. The vendor also quotes a measurement uncertainty associated with the measurement. Our Ireland facility is taking the measurement uncertainty and adding that to the tolerance and then saying its out of tolerance.

I told them that the purpose of the measurement uncertainty is just that - you are not sure of the exact true value. You would need to perform some type of risk associated with the measurement to detemine if it would impact the process.

Thoughts?
 

Jerry Eldred

Forum Moderator
Super Moderator
Re: Measurement uncertainty & tolerance

I won't take any side on the issue, but I can see some possible logic as I'll explain in an example below.

To keep things simple, I'll use "PERCENT OF TOLERANCE" (i.e.: 100% of tolerance is the tolerance limit, etc.).

EXAMPLE: If As Found condition on an instrument was near the tolerance limit (let's say 95% of tolerance), and the measurement uncertainty was (what worked out to be) 10% of that tolerance; that means the As Found (Received) condition could have ranged from 85% to 105% of Tolerance. I hope I'm making sense (and not wanting to ramble too long).

Therefore, an instrument that came back with Received condition In-Tolerance (with the above example) could be POSSIBLY construed as Out-Of-Tolerance. I've thought about this myself in some instances over the years.

However, I would also agree that In-Tolerance is In-Tolerance. I would (personally) not call the item above Out-of-Tolerance because per measurand on the certificate, it was not. But as in my example, if there was a need to be prudent, there may be justification for risk assessment in such borderline cases.
 
J

jfgunn

Your basic question is as follows: What decision rule will you use when taking measurement of uncertainty into account when making pass fail statements?

As an ISO 17025 accredited calibration laboratory (accredited by L-A-B which is part of the ILAC MRA), we have to have a set of decision rules that we use if we make pass/fail statements.

There are two documents that I find helpful is describing how uncertainty should be taken into account.

ASME "B89.7.3.1 - 2001 Guidelines for Decision Rules: Considering Measurement Uncertainty Determining Conformance to Specifications". This document costs $30 from www.asme.org

ILAC "G8:1996 Guidelines on Assessment and Reporting of Compliance with Specification" this is available for free at
http://www.ilac.org/documents/g8_1996.pdf

Note that though the ILAC document is free, I find the examples in the ASME to be a little more descriptive.

Note that there are also facilities that use the opposite approach. They expand the acceptance window by a value equal to the uncertainty. Either approach is reall fine as long as the quality sytem defines what is to be done. It is all a matter of risk.
 
M

merrick65

Re: Measurement uncertainty & tolerance

I will give an example similar to what Jerry posted. I sent a set of gage blocks out for calibration and was reviewing the certificate for errors when I noticed one of the gage blocks was out of tolerance. When I called the calibration source to ask why it wasn't flagged/marked they replied that due to their uncertainty they could not say the block failed and would not mark it as being out of tolerance. The calibration source is accredited and also the manufacturer of the blocks.

I realize this is basically the opposite of the above posted question but thought it was interesting because the same rational is being used in both cases.
 

CalRich

Involved In Discussions
Re: Measurement uncertainty & tolerance

I will give an example similar to what Jerry posted. I sent a set of gage blocks out for calibration and was reviewing the certificate for errors when I noticed one of the gage blocks was out of tolerance. When I called the calibration source to ask why it wasn't flagged/marked they replied that due to their uncertainty they could not say the block failed and would not mark it as being out of tolerance. The calibration source is accredited and also the manufacturer of the blocks.

I realize this is basically the opposite of the above posted question but thought it was interesting because the same rational is being used in both cases.

That's right. The uncertainty is a bilateral value, and so in any particular example, you can say that the uncertainty brings it "in tolerance" or "out of tolerance". As jfgunn alluded to, labs have to determine their decision rule about how they use uncertainty to decide whether these results that are 'on the border' are handled. For those companies that contract a lot of accredited calibrations out, they should likewise have some kind of policy or method to handle results that are on the edge and measurement uncertainty. Essentially, the guy (or gal) who is checking in the equipment needs (and wants) some clearly defined way to know whether they can accept the calibration or not.
 
L

lee01

Often get this one, though not in the Lab environment.

Have had so many discussions with Productive thinking employees who try to take on the basics of MSA and somehow use the percentage of error as extra tolerance, in order to get products out the door.

Yet in the same argument I have the inspectors saying that it should be used to reduce the design tolerance!!!

But I think to myself is it as clear cut as that? Is it more a grey area and not black and white (either add or subtract from the design tolerance). If it includes things like user/ environment etc etc, then how can it be simply add or subtract whatever the percentage from the design tolerance.

I’ll have to get hold of those standards you mentioned, like I said I’m not talking about a Lab environment but I would like to view the thought process of how you introduce the decisions


I could debate this one all day
 

BradM

Leader
Admin
Also, if I'm not mistaken, uncertainty is reported within a confidence level. Thus, it's not a slam-dunk, 100% given the reported uncertainty is your uncertainty, but rather an estimated uncertainty.

BTW, if this instrument is so close, I believe I would be decreasing the calibration interval, suggesting a different tolerance, etc.

Saying, rather you report in/out, the more pressing issue to me is figuring out why it's doing it, and how to keep this issue from happening again.
 

Hershal

Metrologist-Auditor
Trusted Information Resource
Hi everyone,
I have a question regarding a practice that our Ireland facility has been doing with measurement uncertainty.

For example, they send out an instrument for calibration. It comes back stating that the tolerance was within, but it was right at the edge. The vendor also quotes a measurement uncertainty associated with the measurement. Our Ireland facility is taking the measurement uncertainty and adding that to the tolerance and then saying its out of tolerance.

I told them that the purpose of the measurement uncertainty is just that - you are not sure of the exact true value. You would need to perform some type of risk associated with the measurement to detemine if it would impact the process.

Thoughts?

In a pure sense that actually is what MU is for, but it is also a trap.....

A trap because it may suggest that you are OOT when in fact you may not be, but because of the MU, you can't prove it either way.....the other side of that coin, you can't prove it is OOT either.....it casts doubt on the result.....

The obvious easy fix is bring the tolerance closer in, and sometimes there is value in that, but it is not a universal fix.....

One other potential that is now starting to come into vogue as a result of ANSI/NCSL Z540.3-2006 is guardbanding to get a more true picture of the effect of MU and its effect on tolerance.....and I suspect it will be debated till the cows come home, but it may provide one avenue to help your Irish colleagues.....

However, remember they do not work under the American National Standard.....

They are taking the more conservative and safer route.....the best thing they can do is if their stuff can be adjusted, to have it adjusted to nominal.....not just in tolerance.....

Hope this helps.
 
D

Daniel Walker - 2011

Mass calibrations work like this..... acording to NIST and NVLAP....

If the error of the device plus the expanded uncertainty equals an out of tolerance condition, then the device fails.

Calibration labs are also required to set a percentage level of error + uncertainty which warrants adjustment.

Ours is set at 80%. If the error of the device plus the uncertainty exceeds 80% of the tolerance level, then the device must be adjusted back towards it's nominal value. This ensures that the device will not creep out of tolerance shortly after returning to service.
 

Charles Wathen

Involved - Posts
Thanks everyone for your input on this subject. After reading some more, I like Hershal response.

Remember that my calibration department is internal to a company, and we do not perform calibrations for anyone else except internally.
 
Top Bottom