Cal labs for 3458A and their uncertainties

[branadic]

Hi everyone,

what are the uncertainties of your (local) cal labs for 10V and 10k?
The lab in Böblingen (Germany) states:

10V: 2.7 ppm
10k : 7.4ppm
Fluke 5720A

The PTB stated 3ppm for their 3458A on the Maker Fair 2018 for 10V range, as far as I could find on www.mikrocontroller.net.

Prema Semiconductor in Mainz (Germany) states:

20V range: 10ppm
20k range: 20ppm
Datron 4708 & 10K ESI SR 104

Keithley Instruments in Germeringen (Germany) only states a (total?) uncertainty ratio (TUR) between calibrator and DUT of 4 or higher. Deviations from that are given in the protocol and this is in case of a Keithley2002:

20V range: 3.75
20k range: 1.26
Fluke 5720A

-branadic-

[dl1640]

would depend on how it is calibrated, here local Keysight (Shanghai) calibrate our 3458 by comparison method, i.e. use 5730 as source and *Gold* 3458 as reference.
e.g. 5730 output 10v, *Gold* 3458 read 10.000011v, then they calculate test limits based on 10.000011v, not based on a cardinal 10v point.
i remeber the uncertainty for 10v is probably arund 3ppm.

i also they they have almost identical cal system setup around the world so the uncertainty would not differ too much.
unless you have your 3458 calibrated directly compares to PRIMARY standards you get lower uncertainty.

[e61_phil]

Our local cal lab calibrates our 3458As:
10V 4.2ppm
10k 8.5ppm
95% uncertainties. calibrated with a freshly artifact calibrated Fluke 5700A


Our Fluke 8508A was calibrated at Fluke Norwich:
10V 0.7ppm
10k 2.2ppm

These are 99% uncertainties. For 95% it would be even lower. They use a characterized Fluke 5720A.


A couple of days ago I compared 10V on a 3458A with the 8508A and they differ by only 0.2ppm.

[Moon Winx]

Our local cal lab calibrates our 3458As:
10V 4.2ppm
10k 8.5ppm
95% uncertainties. calibrated with a freshly artifact calibrated Fluke 5700A


Our Fluke 8508A was calibrated at Fluke Norwich:
10V 0.7ppm
10k 2.2ppm

These are 99% uncertainties. For 95% it would be even lower. They use a characterized Fluke 5720A.


A couple of days ago I compared 10V on a 3458A with the 8508A and they differ by only 0.2ppm.

Regarding the 99% uncertainties, I believe Fluke tests the items to the 99% test limits (the only limits they guarantee) for pass/fail. The uncertainty of that measurement, however, is probably reported at 95% CL. This reported uncertainty value does you little good though, as the 57xx doesn't have long-term stability specs (anything over 24 hours). The only specs you could use after the calibration is the 90/180/360 day published specs @ either 99 or 95%. 

[martinr33]

So... 4 ppm is also the 3458a spec after the first 24 hours. Do you think the lab is just guaranteeing that spec?

Even if they could get to 0ppm, the unit would only be guaranteed to 4 ppm the next day anyway (or 2ppm with the high precision option).

[dl1640]

the spec of your 3458 will not change,
the reported uncertainty is only given for a specific test setup at the time the calibration is carried out.
it is largely affected by the uncertainty of the cal standards and short-term behavier of the 3458 under test,
so, lower uncertainty means better cal standards and/or technique used.

the spec of 3458 contains many things, especially long-term drift, tempco etc., thats why it is often specified at as a time period, e.g. 90days, 1year, with a certain temperature range e.g. 18~28 degC, along with other conditions.

[e61_phil]

Regarding the 99% uncertainties, I believe Fluke tests the items to the 99% test limits (the only limits they guarantee) for pass/fail. The uncertainty of that measurement, however, is probably reported at 95% CL.

Why do you think this test has only 95% CL?

This reported uncertainty value does you little good though, as the 57xx doesn't have long-term stability specs (anything over 24 hours). The only specs you could use after the calibration is the 90/180/360 day published specs @ either 99 or 95%.

Can't follow you. What is the point with the 57xx? Aren't we talking about the 8508A?

the spec of your 3458 will not change,
the reported uncertainty is only given for a specific test setup at the time the calibration is carried out.
it is largely affected by the uncertainty of the cal standards and short-term behavier of the 3458 under test,
so, lower uncertainty means better cal standards and/or technique used.

It might not improve anything for the 3458A, but the 8508A has also specs relative to the cal standard. And in this case you have a benefit with lower calibration uncertainties.


Does anyone know why you have to add this 2ppm for NIST traceability for the 3458A? In some publications they compare the 1-year specs of the 3458A and 8508A and if you add everything from the datasheet, the 8508A specs are more than twice as good as the 3458A for 10V (10.05ppm vs 3.4ppm)

[dl1640]

the 57xx do have *relatively* uncertainty spec,
if one have better cal standards than fluke have to calibrate 57xx,
then the *abosulte* uncertainty of 57xx could be calculated as new spec, better than fluke published one.
but the published *relative* spec contains many things...it is defined for long-term behavier.

i think 2ppm is the uncertainty transfered from NIST 10v to keysight 10v (NIST 732 to KS 732?? maybe),
but anyway 3458 spec has a lot of head room, if one has years of calibration history of 3458, i think it is common that it is not out of 1-year spec for years, and so is fluke 8508 i think.


so a good calibration and history is really helpful for one to estimate a better spec rather than published 1-year spec, that calibration make sense.
with lower or higher cal uncertainty it is for one to trust the reported cal data, more or less. normally TUR is kept below 4:1 or 3:1 or use transfer technique, like a specially calibrated 57xx to calibrate a 3458, the drift and repeatability is thoroughly verified at *short* period, e.g. every 30days.

[Alex Nikitin]

On the calibration certificate for our 3458A from Keysight UK the uncertainties stated are 3ppm for 10V (3.1ppm for 1V) and 7.5ppm for 10K .

Cheers

Alex

[dl1640]

https://www.eevblog.com/forum/metrology/the-calibration-of-hp-3458a-_gold_-version/msg1383836/#msg1383836

10V: 2.8ppm (@ KS Shanghai)

KS network has similar uncertainty around world.
the *gold*3458 is calibrated at US headquarter, and 5730 is calibrated at local NMI, so uncertainty may differs at each local KS cal lab due to 5730 cal uncertainty i suppose.

[ap]

There are a few things to consider re uncertainties. First the uncertainties of the standards used to adjust the 3458a. These are (in a lab suitable to adjust the 3458a) normally a set of voltage standards (732...) and a low drift 10k standard (sr104 or the like). Using a 5720 as the standard would not be good practice (uncertainties). These standards can be maintained at 1ppm uncertainty or better. Even with the 2 / 3 ppm that is specified by keysight, the 10v and 10k are then at about 6 and 11.5ppm at 90 day interval limits. This is the spec data that should be used for verification. 24 hours data is a time problem in most cases, 1year makes no sense, you want to use the meter thereafter using 1 year limits.
It is pretty obvious though, that using verification standards that are at about 4 ppm and 8 ppm as stated somewhere above in examples of cal labs uncertainties is inadequate to verify the 3458a in these ranges. The TUR with these uncertainties is just insufficient. And if you have an option 002 meter, it even gets worse at 10v

[e61_phil]

normally a set of voltage standards (732...) and a low drift 10k standard (sr104 or the like). Using a 5720 as the standard would not be good practice (uncertainties).

But with 10V and 10k only, one cannot calibrate a multimeter. DCV might be ok with a calibrator a Fluke 752 and Fluke 732, but the rest is more tricky.

[dl1640]

i think to calibarte 3458, a 5720/5730 acts like a *transfer*,
such a calibrator is characterized for each test point sufficient for full verification of 3458.
the way to calibrate the calibratr is by comparing to reference 8508, various dividers, resistors/shunts, dcv reference etc.
these standards are calibrated using high-end techniques in primary lab.

the 5720/5730 is verified at regular basis, e.g. 30days or 90days, and upon lots of history data, it might be well wthin 24hrs spec in any time.
because they are *constantly* calibrated, but the actual traceability goes to national standards: dcv, ratio, resistance...

not every lab can prepare such a 5720/5730.

i think 3458 manufaturing line calibrate using 5720/5730 too.

[EEVblog]

[e61_phil]

Calibrating the Hewlett-Packard
3458A DMM with the
Fluke 5720A Multifunction Calibrator
https://content.fluke.com/comx/applications/3458.pdf


Characterizing & Disciplining Electrical Calibrator Instrumentation to Improve Test Accuracies & Measurement Uncertainties
http://download.flukecal.com/pub/literature/ImprovingTURs2007ncslpaper.pdf


Calibrating Precision Multimeters Using a Characterized Multifunction Calibrator
http://download.flukecal.com/pub/literature/gbennett_ncsl_2004.pdf

[ap]

But with 10V and 10k only, one cannot calibrate a multimeter. DCV might be ok with a calibrator a Fluke 752 and Fluke 732, but the rest is more tricky.

The 10v and 10k are in the first place intended for adjustment only. Verification (=calibration) is a different issue, and if e.g. one takes a standard 5720 at 10mA, its spec is about 110ppm (90 days), while the 3458a is about 20 ppm. Characterizing the 5720 current uncertainty may help a little, and only if the lab does it.
So using a pecision calibrator is always a compromise. It is convenient and fast, and thus often less pricy, but using separate standards for resistance and current (via precise shunts) leads to lower uncertainties.
So when having a 3458a adjusted, one should first look at what uncertainties the cal lab specifies for the 10v/10k standards and secondly for the verification. The fact that a cal lab states it will calibrate a 3458a does not mean a lot. I have seen labs using an HP 8903 to calibrate an 8 digit ac calibrator...

[e61_phil]

Verification (=calibration) is a different issue, and if e.g. one takes a standard 5720 at 10mA, its spec is about 110ppm (90 days), while the 3458a is about 20 ppm. Characterizing the 5720 current uncertainty may help a little, and only if the lab does it.
So using a pecision calibrator is always a compromise. It is convenient and fast, and thus often less pricy, but using separate standards for resistance and current (via precise shunts) leads to lower uncertainties.

Verification is (hopefully) much more often needed than adjustement. Especially for the 3458A where only artifcat calibration is possible, which is really a pity.

I had a look into the 8508A cal cert. They stated 4.3ppm uncertainty for 10mA. And they did it with the 5720A, because there is no other meter in the list of used standards.

[ap]

Well, first of, the 5720a 90 day spec is 30ppm plus 40nA, not 40ppm of range, so i read that wrong, total 10mA spec is thus 34 ppm, for 24 hours it is 29 ppm. So either they did some "characterisation", with 4.3 ppm being pretty much on the extreme side I would think, does not sound very plausible for "characterization", so i would rather think they used the 5720 as the source only and another meter (8508a ?) for measurement.

[dl1640]

to calibrate full test points of 3458, it is good to use a characterized calibrator, because the 3458 is only tested to within its 1-year spec.

if one need much lower uncertainty, discrete standards do the job, but much more time/labor consuming.

only 10v and 10kohm standards are just able to compares to 3458 internal reference for *artefact* adjustment.


if one have a significant amount 3458 to check, a characterized 5720/5730 is just fine.


local KS service center does not have a characterized 5730, so they use a verified 5730 (i guess to verify it to satisffy 1-year spec at NMI) and a *gold*3458, and maybe some swith box is used for cabling change and that switch and cabling effects are compensated or minimized.
so first *gold*3458 measures, then uut 3458 measures, or vice versa.

[dl1640]

fantastic video and tour to primary lab,
just can't get enough, want more

[e61_phil]

Well, first of, the 5720a 90 day spec is 30ppm plus 40nA, not 40ppm of range, so i read that wrong, total 10mA spec is thus 34 ppm, for 24 hours it is 29 ppm. So either they did some "characterisation", with 4.3 ppm being pretty much on the extreme side I would think, does not sound very plausible for "characterization", so i would rather think they used the 5720 as the source only and another meter (8508a ?) for measurement.

I fully agree, it is hard to believe. Even the stability spec for 10mA is worse than 4.3ppm. But there is no other instrument in the cal cert.

- Fluke 5720A
- Fluke 5725A
- Fluke 8508A-7000k (that is only a set of cables)
- Fluke 742A-1
- 10G calibration piece

That's all. No golden 8508A or something like that.

[ap]

Yes, the daily drift (4h, within 1°C, 95%) is about 24ppm at 10mA. Thats a factor of roughly 5,5 worse than the spec of the stability applied in your cal. Stability over that time is 10ppm. Factor 2+.
A 5720A app note calls for using a 742 current shunt, using the 5720 as a source, and using the already calibrated 3458A (in DCV) to measure the current through the shunt method. They use the short term stability (24h) of the 5720. They end up with a composite uncertainty of 15,9ppm (factor 3.7 worse than what the cal lab used). Sure one can use a better spec for the 3458A DCV uncertainty and the 742A stability/uncertainty, but the 10ppm stability spec of the 5720A remains. A 742A or thelike with suitable value is not listed in the test gear list however.

So the question remains, how did they end up with such a high factor of improvement. It remains very mysterious. Could be asked...

Interesting is that the 5720A app note uses 3458A verification uncertainty limits of its 1 year spec. That is nonsense for any practical purpose. If a meter is just within limits after cal, how should the customer use it for another year beyond the cal time (using 2 years spec then?; that gets even worse...). 90 day limits should be used. So the app note indirectly highlights the limitations of the 5720A in calibrating the 3458A. Also the 1 year 3458A uncertainty is listed with 36.2ppm in the Fluke app note , not what I read in the datasheet (mine says total of 30ppm using the not very good Keysight cal standard uncertainties of 2+3ppm @ 10V/10k), but certainly helping in getting a better TUR in the app note...

[dl1640]

Calibrating Precision Multimeters Using a Characterized Multifunction Calibrator
http://download.flukecal.com/pub/literature/gbennett_ncsl_2004.pdf

' For this paper the specifications shown are absolute 1 year specifications at 99 % confidence level. I added the gain and floor together [2, 3, 4]. For the Agilent 3458A/HFL the additional error for Agilent factory traceability was added to make these specifications absolute. The uncertainty limits shown in the charts, labeled “Char Uncert”, are uncertainties assigned to the characterized calibrator and are at the 95 % confidence level. Uncertainties reported for the calibration of the DMMs include the uncertainty of the characterized calibrator, Type A components of the DMMs and in some cases additional adders for system effects. The reported uncertainties for the DMMs are not provided in this paper. '

their (fluke) idea is to verify the longterm behavior of 5720A, for 12 mo in this paper, to get a better spec than published 24 hr one.
such characterized uncertainty of the calibrator is at 95% confidence level.
the reported uncertainty for the calibration of the DMM 3458A/8508A is unknown by this paper.
if the TUR is not enough, a guardbanding will come, i think.

[mzzj]

Regarding the 99% uncertainties, I believe Fluke tests the items to the 99% test limits (the only limits they guarantee) for pass/fail. The uncertainty of that measurement, however, is probably reported at 95% CL.

Why do you think this test has only 95% CL?

Quoting the calibration uncertainties at 99% level sounds weird when most of the stuff is quoted at K=2 95% level but I'm too lazy to check what ISO 17025 has to say about it.
(stability specifications are often quoted at some other level)

Anyways Fluke Norwich calibration uncertainties are really good as their UKAS CMC is 0.27ppm for 10 volts (K=2, 95%)
https://us.flukecal.com/literature/accreditations/norwich-fluke-precision-measurement-ltd

We use local national metronomy  lab for your 3458A, 1ppm for 10 volts.
Edit: local NML also uses 5720A and 3458A as reference equipment but they can calibrate those in-house against their Josephson standard and quantum hall resistance referecence.
I'd suspect that Fluke Norwich also has Josephson volt standard and QHE so they don't need to rely much on 57xx stability

[mzzj]

Both Keysight and Fluke in US have their own Josephson standards, apparently Fluke Deutschland also has one and guessing from uncertainties I'd say that Fluke Norwich also have their own.
NIST in US obviously has their own also but US Navy had more of them than rest of the primary labs combined. Apparently they had couple of millions extra left before end of accounting year that had to be spent somewhere 

How and Why: http://assets.fluke.com/appnotes/calibration/josephson-voltage-standard-in-a-working-calibration-laboratory.pdf

[e61_phil]

Quoting the calibration uncertainties at 99% level sounds weird when most of the stuff is quoted at K=2 95% level but I'm too lazy to check what ISO 17025 has to say about it.
(stability specifications are often quoted at some other level)

I thought (without any evidence) Fluke sticks to their 99% levels. The old instruments (5440B, 5450A..) are only specified at 99%. Only the "newer" ones has additional 95% specs to be comparable with other equipment.

Both Keysight and Fluke in US have their own Josephson standards, apparently Fluke Deutschland also has one and guessing from uncertainties I'd say that Fluke Norwich also have their own.

The next calibration of our 8508A will take place in germany, I think.

[mzzj]

The next calibration of our 8508A will take place in germany, I think.

https://www.dakks.de/as/ast/d/D-K-15123-01-00.pdf
Looks to be "good enough" at 0.2ppm for 10volts

[mzzj]

And one way to look for ISO 17025 certified labs is from the ILAC-MRA website:
https://ilac.org/signatory-search/

Select for example Germany and you get link to DAkkS website with a list of all accredited calibration labs in Germany. Still takes some time to skim trough their CMC (calibration and measurement capabilities) "scope"
When you select accredited lab you can also trust to the results more... unless its US lab. Those goddamn luddites are still not able to grasp the idea of "total calibration uncertainty" even if they are accredited 17025 lab 

[Henrik_V]

...
The PTB stated 3ppm for their 3458A on the Maker Fair 2018 for 10V range, as far as I could find on www.mikrocontroller.net.
...

-branadic-

NO.... never ever ..  not at the fair.  The 3458 had been calibrated UNDER LAB CONDITIONS to the cited 3ppm.  They might have performed better, but don't deal serious ppm at a fair  ... thats not faire  äh serious. 

For uncertainties there are databases the CMCs at the BIPM for the national laboratories and at  DAkkS for the german ISO 17025 labs. Same can be found for other nations.  Just keep in mind that the stated uncertainties migth not be the 'standard' uncertainties you get for the standard price
 

[mzzj]

  Just keep in mind that the stated uncertainties migth not be the 'standard' uncertainties you get for the standard price
 

Much true.
And the CMC values are sort of "ideal case" when the meter under calibration is enough good/stable.  You can get  different uncertainties for two different 3458A meters from the same lab.

[Moon Winx]

Quoting the calibration uncertainties at 99% level sounds weird when most of the stuff is quoted at K=2 95% level but I'm too lazy to check what ISO 17025 has to say about it.
(stability specifications are often quoted at some other level)

I thought (without any evidence) Fluke sticks to their 99% levels. The old instruments (5440B, 5450A..) are only specified at 99%. Only the "newer" ones has additional 95% specs to be comparable with other equipment.

Both Keysight and Fluke in US have their own Josephson standards, apparently Fluke Deutschland also has one and guessing from uncertainties I'd say that Fluke Norwich also have their own.

The next calibration of our 8508A will take place in germany, I think.

It's confusing, but maybe this will help clear it up.

Fluke uses 99% CL in their manufacturing quality control and so the guaranteed specifications are presented at the 99% CL (for a great explanation of this, see http://download.flukecal.com/pub/literature/msc04.pdf).

The 95% specifications are listed due to contract obligations from a specific organization that requested published 95% specifications as a requirement in order to buy the calibrators.

Don't use the 95% specifications unless your supporting lab verifies the calibrator to the 95% specifications. Why? Because if you have a unit that is out of tolerance and needs repair, you may send it to Fluke, who, by verifying it at 99% specs, may not see a problem.

The confusing part is that labs use measurements with uncertainties presented at 95% to compare to Fluke's 99% test limits to determine pass/fail. It's confusing until you are convinced of the fact that your measurement uncertainty CL has nothing to do with the CL of the specifications you are comparing against.

If Fluke's 95% published specs were proportional to the 99% by a constant (they are not), you *could* use the specifications @ 95% even when tested at 99%. But until someone from Fluke clears this up, I"d avoid the 95% listing altogether.