PMEL Forum

One suggestion I would make....

I do most of my uncertainty work in Excel.  I would highly recommend that unless you go get training in one of the freeware or paid tools, that you take some classes in Excel.  There is a lot of free information about Excel on the internet.  The first prerequisite I think is that you need to understand the math under uncertainty calculations before learning Excel.  You need to learn what the formulas are, how to do the statistics and other functions.  Then do some searching for tutorials for Excel on how to do those formulas.  You will find, I think, that if you know how to do the math, Excel makes it a lot easier.  I was self-taught in both Excel and uncertainty.  I did tons of web searches and found many power points and pdf's from various organizations with breakdowns of what math is involved. 

There is no shortcut.  You will have to learn all the math (most of it can be done with Algebra.  The statistics can be done by Excel using simplified formulas).

We use the Quametec uncertainty spreadsheets and it is easy to use and presents everything for an auditor. I am starting to wonder if all of this uncertainty calculation is really worth the investment, though. We are making the exact same measurements using the exact same standards. The only difference is I get a little better idea of how good the measurement is. With all of the investment in software and time to perform uncertainty budgets, where is my ROI? Sure I can charge my customer more for the additional calculations, but how does that serve them better? I am usually giving them more data on top of data that many of them don't understand anyway. I can understand providing uncertainty from a higher tier lab to a secondary lab, but for the end user, I don't get it.

I agree.  We have customers doing work in contexts where they need accredited cals on everything, in many cases where they don't need, use or even understand the uncertainty calculations.  They just know they need to have them.  Costs a lot of effort doing calculations that aren't used.

I think the same goes for providing data most of the time to customers.  Some auditor told them they needed to have it and they never end up even utilizing it!  Unless its a fixed value and/or primary standard why would you care to know exact values for all the ranges and functions for general purpose equipment?

That's where your auditor would be concerned.

Now, to speak out both sides of my mouth...

I have been in the cal world for a very long time, since the MIL-STD-45662 days (the version without the A at the end).

It has taken me a long time to come around.  But I am gradually getting "brainwashed" in this.  After going kicking and screaming for many years, I do get it.  A NIST traceable (or other traceability) measured value is only traceable if there is an uncertainty associated with it.  From a basic calibration technician perspective, or a user perspective, it doesn't seem to make sense.  You use a good calibrated standard with a "4:1 TAR" (I know it's an outdated term - here for illustration).  But from a metrologist/metrological perspective, if you measure a UUT, by good metrological definition, you need to know how good the measurement is.  If you don't, it's not really a valid measurand.  So although posting MU on certificates is of little use to many lower echelon users, it is by definition, a part of what makes it a valid value.

There is, after all, in any context where the value and how quantitatively accurate that value is, the by definition need to know that it validly meets the measurement requirement.  Even an end user of a Fluke 87 measuring 115 VAC at 60 Hz needs to know that the accuracy of that measurement meets their needs.  If for example you "calibrated" it by plugging a light bulb into an outlet and the tolerances for the lightbulb were UTL=blow out the bulb, and LTL=bulb does not light up (disgustingly loose for illustrative purposes), and you then used that as your standard for "calibrate the DMM, that would not be adequate.  Even for the least accurate application of that DMM, there is some uncertainty of measurement needed, even if the user doesn't recognize or use GUM principles.

So, I give in.  Calibrations need uncertainties, period (in some form).  Because a measurand is not a measurand without them.  There, I spoke out of the other side of my mouth.

wow!! half of that is so wrong!

There is no need to report them to people who won't use them.  But there is the need for them to exist, reported or not.  This is speaking philosophically/theoretically.  That is my point, that when you make any quantitative measurement, it is only of value when there is at the least, some implicit understanding that the measurement is good enough for its intended purpose.  That is what I think is misunderstood about the entire topic of MU.  When you use a Fluke 87 to measure a voltage, there is the implicit need to know that it is a "good" value.  My illustration was that if I am measuring 115 VAC and the extent to which I know it is "good" is that even if the DMM reads 115, if I only know that the accuracy of the measurement is between 0 and infinity, then that 115 V reading is meaningless.  For every tolerance, for traceability, there must be an uncertainty.  There always is an uncertainty.  Whether someone calculates it in accordance with GUM is a separate topic.  But no measurement is traceable unless there is an associated uncertainty (reported/calculated or not). 

One of my favorite papers that discusses the need for measurement uncertainty is NPL Good Practice Guide no. 131, "Beginner's Guide to Measurement in Mechanical Engineering" from the National Physical Laboratory in the UK. It treats the subject in layman's terms, and even has the famous exercise where you build an uncertainty budget for the length of a piece of string.
http://www.npl.co.uk/publications/guides/guides-by-number/

I use the Quametec Excel add-in from time to time, but it can get a bit complicated (after years of NCSLI conference tutorials on the subject, things like Student's T and Welch-Satterthwaite Formula still fly over my head). I finally came up with a simplified version of my own, based on the Excel sheet from NRC-CA, mentioned in the first reply:

Quote from: OlDave on 10-29-2015 -- 16:44:41
www.nrc-cnrc.gc.ca/obj/doc/solutions-solutions/advisory-consultatifs/clas-clas/uncertainty_budget_template.xls

I validated it against the above mentioned programs, as a sanity check, and even our 17025 accreditation assessors have reviewed my uncertainty budgets and have given a nod to them.

Quote from: Hawaii596 on 11-06-2015 -- 09:17:40
For every tolerance, for traceability, there must be an uncertainty.  There always is an uncertainty.  Whether someone calculates it in accordance with GUM is a separate topic.  But no measurement is traceable unless there is an associated uncertainty (reported/calculated or not).

Well said.  I would only add that there also needs to be an unbroken chain of measurements, and a quality system implemented to ensure the associated measurement equipment maintains their measurement uncertainty.

whatever happened to the 4 to 1 for instruments and 10 to 1 for standards rule. This still works, but there was no money to be made. Hence, uncertainty budgets.It's all a brainwash to me. If I calibrate a sig gen with an 8902A system that was calibrated at Keysight, there shouldn't be any uncertainty budgets needed. It's all a ruse to get you to shell out bucks.Nothing but BS!!

The other problem is nothing is standardized. It depends on who is your accrediting body. That will determine how you do things.
I have just completed my 5th inspection by A2LA and all was in order.Some companies, such as K---, shop around for the least intrusive inspection body so they can do things their way.They were with A2LA, then went to A-Class and now LAB. Using the same equipment,their Uncertainty Budgets are way better than mine.Just a bunch of BS!

Ok. Сan you explain me, how calibrate Fluke 8508a with Fluke 5720 on DCV mode. TUR ~1:1. And when i can use 24h, 90 day specification?

We just finished an Fluke 8508A Procedure for a customer.   
You don't calibrate the 8508A with the 5720A.  You have to characterize it with your labs best standards, and you have to have solid data behind your standards.  Then the Fluke 5720 is used as a transfer standard.
For DC Voltage you would have to use a 10V cell, an accurate / known divider network and a null meter.  Then you can use the 5720's 24 hr Specs.

Mike 

Quote from: CalLabSolutions on 11-08-2015 -- 22:05:05
We just finished an Fluke 8508A Procedure for a customer.   
You don't calibrate the 8508A with the 5720A.  You have to characterize it with your labs best standards, and you have to have solid data behind your standards.  Then the Fluke 5720 is used as a transfer standard.
For DC Voltage you would have to use a 10V cell, an accurate / known divider network and a null meter.  Then you can use the 5720's 24 hr Specs.

Mike

We have DC Fluke 732b-10V, divider Fluke 752a, and null indicator Keithley 2182. With this units i calibrate 5720 (100 mV, 1V, 10V,100V, and 1kV range), and than i can use specification 24hr+this U95 calibration certificate in uncertainty budget?