Uncertainty and Resolution of Standard

[Hawaii596]

I am not an uncertainty expert.  So I'd appreciate if someone knowledgable might be able to assist.  If for example, you are using a Fluke 5720A to calibrate an HP 34401A; if you are making a measurement at 10 VDC....

I understand the resolution of the 34401A is part of the uncertainty of the measurement.  But I fail to understand how the resolution of the 5720A contributes to the uncertainty.  I know some accreditation bodies define the resolution of the calibrator as a contributor, but others do not.  Is there a universal answer on this?  My logic is that if you enter 10 VDC as the output from the calibrator, resolution for the calibrator (in this example, Fluke 5720A) does not apply, as it truly isn't used.  I can understand resolution can be a contributor if it forms a part of the measurement (such as the DMM resolution in this example).

Any quick inputs appreciated.

[griff61]

If you look at A2LA G110 - Guidance on Uncertainty Budgets for Electrical Parameters, it essentially states that you must consider the resolution of your standard, but you don't necessarily have to include it in your budget if its contribution to total uncertainty is negligible and has essentially no effect.

http://www.a2la.org/guidance/A2LA_G110.pdf

hope that helps.

[Hawaii596]

Not to opine on either AB, we use ACLASS, which doesn't have that specificity in the guidance document.  Another associated quality person is using the A2LA guidance document - and I was looking at the requirements in it last night.  I'm not uncertainty expert, and I plug along and do the best I can with my level of knowledge.  So I am left to my instincts in combination with the knowledge.  It could very well be that "the right" way to do this is always include the resolution of the standard.  I think because I am so instinctive in how I do things (I'm not good at memorizing and following rules - how did I ever survive ten years in the Navy?).

So when I instinctively look at resolution of the standard where I "instinctively" don't understand it, I struggle a little.  In an early morning conversation on the topic today, I theorized if I used a Fluke 732B 10 VDC reference to cal 10 VDC on an HP 34401A, on the 34401A and cabling there are the regular suspects (repeatability, resolution, etc.).  And on the 732B there is the uncertainty of the value.  The uncertainty of that value has imbedded in it (I think) and applicable resolutions involved with its derivation.  If I then do the same cal measurement with a 5720A, set it for 10 VDC, I just can't wrap my aging brain around why the resolution becomes important.  It seems like the specs of the 5720A and the uncertainty from the certificate of calibration for it captures that between them, and that adding them in once again to budgets is duplicating.

I guess for the sake of keeping things simple, making such a rule for those times when it does make a difference, and doing it the same way all the time makes your budgets a little more conservative.  So it does create some buffer.

The point I think I'm at in processing this is in agreement with griff61's comment thatyou must consider resolution of the standard, but don't have to include in your budget if negligible and has essentially no effect.

What I'm still trying to figure out is the underlying theory.  I understand the reason the LSD (least significant digit) on a meter is of consequence is that when you apply a value it can toggle between a couple of LSD counts, and because there is no further resolution, you have to call that an uncertainty contributor.  But I guess it just seems to me that doesn't apply to the source in a calibration step.  If I set 10 VDC (lets call it 10.000000 VDC = 1 uV resolution), there is no "toggling" between LSD's.  It is a setpoint with an uncertainty and a tolerance that is completely unimpacted by that 1 uV decimal.  I don't read that 1 uV character in the calibration.  Even if I have a setpoint on the calibrator of 10.000001 VDC, that digit does not "toggle" and cause potential error in the calibration.  The repeatability, uncertainty, tolerance, environment, cabling, etc. all do - but that displayed digit doesn't.

Again, not trying to be contentious.  I am about to get into a debate with other quality people in my organization about this.  If I'm wrong, please help me understand.  I honestly feel like due to my lack of learning on the topic I could very well be wrong.  But I definitely want to be right (not think I'm right - be right).  If that means changing my views on this, so be it.  It seems like such a dumb thing to get hung up on, and I am very grateful for the inputs thus far - thanks to person also who already sent me a couple of PPTs that I am now reading before my meeting.

Thanks all.  Great having a PMEL community that I can share, query and vent with.  Hope you all are having a good new year.

[RFCAL]

I agree with griff. You must show that it is irrelevant before you can omit it. We just include it in all calculations because A2LA prefers to see it.
You can take this one step further---Inspection bodies want to see the spec of the UUT in the calculation. In fact, A2LA has stated to me that this MUST be the most significant contributor. If that is the case, why not just list the specs of the UUT as the uncertainty. You would not be that far off--everything else contributes very minimally. We could then forgo all these ridiculous calculations and regulations. Bring back the old 4:1 & 10:1 standard.

[griff61]

In most cases, when you are doing your uncertainty budget, you are comparing your best source with your best measurement ability, so the uncertainty contribution from the resolution of your source should be negligible. The decision to include it or not comes from the situation where your standard is not as capable, for instance say you were doing a budget for frequency and had to use the GPS signal to check a rubidium standard. (which we once did as a courtesy). Your standard's resolution is less than or equal to your unit under test, so the effect on the uncertainty budget for your measurement would be altered to a greater extent due to the resolution of the GPS.
On the other hand, if you are doing a Fluke 87 with a 5720A, the resolution of the 87 is more critical than the resolution of the 5720A.

If anyone wants to correct me, please feel free, because this makes my brain hurt...

I used the A2LA guidance because it is basically boiling down the information in JCGM 100:2008, GUM 1995 Evaluation of measurement data — Guide to the expression of uncertainty in measurement

http://www.bipm.org/utils/common/documents/jcgm/JCGM_100_2008_E.pdf

Of course, leaving the resolution in there is not really that much additional work if you are using a spreadsheet and if that makes everyone happy, so much the better.

[RFCAL]

yep,agreed. got tired of arguing with A2LA.

[Hawaii596]

Good inputs.  Folks just got here for the meeting about this, and we'll see how it goes.  We don't use A2LA, but the other quality person pulled the A2LA guidance document as evidence for the viewpoint.  But... that said, well put that sounds like the good practice is that you have to either show that it is not significant/negligible or enter it.  And it just keeps the auditors happier when you just include it.  If it is not significant/negligible, and you DO include it, that is also your written evidence that it is indeed not significant/negligible.  I haven't previously included it because it hasn't been made an issue.  But my interpretation of your above inputs at least helps me understand what/why I need to do it.  Thanks again.  Please feel free to add further comments if you think of them, as I am working on this all day, and I will periodically check back.

[Bryan]

In the case of the 5720 I think it's irrelevant because when you set it for 10 volts you set it for 10 volts period, it has dividers that divide by 10, not 9 or 11.  But I do always consider it in the case of something like the 732 with a reported value.  Unfortunately I don't get to make the rules.  Heaven help someone using something like a PG506 to calibrate a scope, the 0.25% output specification would be buried by the resolution.   

[Martin]

If you look at A2LA G110 - Guidance on Uncertainty Budgets for Electrical Parameters, it essentially states that you must consider the resolution of your standard, but you don't necessarily have to include it in your budget if its contribution to total uncertainty is negligible and has essentially no effect.

http://www.a2la.org/guidance/A2LA_G110.pdf

hope that helps.

Is there a definition or guidance for what is/isn't "negligible"?

[griff61]

If you look at A2LA G110 - Guidance on Uncertainty Budgets for Electrical Parameters, it essentially states that you must consider the resolution of your standard, but you don't necessarily have to include it in your budget if its contribution to total uncertainty is negligible and has essentially no effect.

http://www.a2la.org/guidance/A2LA_G110.pdf

hope that helps.

Is there a definition or guidance for what is/isn't "negligible"?

If you look at table 5, the contribution of the standard's resolution to total uncertainty is 0.00007%
In this case, not including the resolution would have virtually no effect on your uncertainty.
Both the GUM and A2LA pretty much say that you should include as many factors as possible. However, the recommendation is still to include it in your budget and to be consistent in what is included in your calculations.
Table 6 lists the type A and B contributors that at least should be considered for inclusion.
In my opinion, including as many as possible, if not all of them, makes it more practical to create a method to calculate uncertainty across a large diversity of equipment with a single worksheet and set of instructions.

You can play with the numbers in the tables and see what that contributor's percentage of influence becomes in the % of total column

[Bryan]

A2LA P110 para 4.1.1 line b) defines significant as 5% or greater. BUT this policy deals with CMC claims to support a scope of accreditation, may not have any bearing on the original scenario of this thread.

[beadwork]

Here's a good resource for Measurement Uncertainty information:
http://elsmar.com/Forums/forumdisplay.php?f=135

Anyone using Quametec's Uncertainty Toolbox?

[Martin]

A2LA P110 para 4.1.1 line b) defines significant as 5% or greater. BUT this policy deals with CMC claims to support a scope of accreditation, may not have any bearing on the original scenario of this thread.

I have heard of 5% being used before, but didn't know where it was in any guidance document or when it could be applied. I figured someone would be able to quote it.

[RFCAL]

We are using the toolbox. Great software and easy to use.Also, I have heard from A2LA that if you include an unc. and it doesn't change your final outcome. by more than 5%, then you can omit it.

[NC-Cals]

I use the Quametec Uncertainty Toolbox to figure uncertainties.  The resolution of the standard does not need to be included in your uncertainty budget IF you use the uncertainty of the calibration of that standard.  This is because the calibration uncertainty of the standard should include the resolution of the standard.  Also, Fluke has a reputation of including all contributing factors in their accuracy specifications.  Other companies are not as forthcoming with their actual accuracy specifications and you need to read their literature closely.

So far, each inspector I have worked with has their own interpretation of contributing factors to an uncertainty budget.  If you go through the steps to prove that a contributor is less that 5%, they are going to want to see your work.  Then they are going to want to see you SOP for calculating uncertainty budgets.  Sometimes it is easier to just include everything unless you like debating with inspectors.  That can be fun, too.