Fluke calibration misconseption - worth knowing!

[jonpaul]

All:

This discussion seems bizzare,  We Have a fine Fluke 8842A  5 1/2 d DVM, the CAL button allows procedures for ADC, DCV, ACV, ohms 2 wire Ohms 4 wire amps.

Every function can be CALed and opon completion of each routine the constants are stroed in an EEPROM or NVRAM.

After the CAL there is a noticeable improvement in accuracy.

Perhaps the lower accuracy DVMs do not have a CAL feature.

Kind Regards,

Jon

[bdunham7]

That's just ass-covering. Does it happen in practice?

I've seen 30-year old meters that agree with brand new ones down to the last digit. This makes me suspect that most calibration certificates are just ass-covering for legal reasons.

(for devices that are based around an internal voltage reference)

Typical performance will be much better than the specified uncertainties from a reputable manufacturer.  That's just the natural result of including the environmental limits and a 95% or 99% confidence interval in the specifications.  You can try and estimate the typical performance from specs from someone like Fluke/HPAK/TeKeithley, whereas specs from some 2nd tier manufacturers may just be aspirational. 

Long-term drift is a real thing, but varies greatly with the device and how it is selected and burned in.  In most non-metrology devices, other factors will outweigh reference drift. 

[bdunham7]

Perhaps the lower accuracy DVMs do not have a CAL feature.

I think most of them have a cal feature, although it may be practically inaccessible in some cases.  Every modern Fluke that I'm aware of--and every one I own that doesn't have trimpots--has a method similar to your 8842A in which you enter CAL mode, select a mode and range and then present the meter with a calibrated stimulus.  Handheld DMMs typically won't allow you to use custom values like the 8842A as this is less necessary with lower precision. 

[Arhigos]

if you check fluke 754 specification: 1 year 3vdc is 0.02 %  and 2 year is 0.03 % .

That's just ass-covering. Does it happen in practice?

I've seen 30-year old meters that agree with brand new ones down to the last digit. This makes me suspect that most calibration certificates are just ass-covering for legal reasons.

(for devices that are based around an internal voltage reference)

Or course it's happening. Why do you think calibration labs even exists lol?

What kind of 30 year old multimeters you had and how you compared them?

[Fungus]

Or course it's happening. Why do you think calibration labs even exists lol?

For verification and issuing of paperwork? 

Where's the example of a meter that went out of spec?

[bdunham7]

Where's the example of a meter that went out of spec?

https://www.eevblog.com/forum/testgear/hp-34401a-reads-a-little-low-on-ac/msg3241784/#msg3241784

[Brumby]

Just my thoughts.......

I will go out on a limb here and say that, unless you are working at quantum levels where the laws of physics ensures unwavering quantities, EVERYTHING drifts.  The question is by how much and over what time period.

If you have a device that drifts two orders of magnitude below its finest resolution, then you will never see it - and, clearly, there are some devices which we have encountered that appear eligible for that category!

The whole purpose of "calibration" is to ensure confidence in the measurements taken.  When you get into metrology level measurements, from what I have encountered, the word "adjustment" sends shivers down spines.  In these situations, the HISTORY of comparison to a standard on how a device performs is at least - if not more - important than the point-in-time accuracy.  Even if a device falls out of spec, if we have a record of the history, then we can still have a high level of confidence in what the actual values were at the time of measurement.

For example, if you have a meter which is used to measure 1V to within a 50nV tolerance and the calibration history (comparison to a standard) shows it read 4nV low three years ago. 8nV low two years ago, 12nV low last year and 16nV low right now, then you can be pretty confident you have a linear change - and that this will allow you to determine the actual voltage at any time between these points with a very high degree of confidence.  Importantly, it will also indicate the expected deviation over the following year.  In the above example, you will have a high level of confidence that the unit will read 18nV low in six months.  You will be keen to see the calibration report in a year's time to ensure this was correct, but in the mean time, you will have reasonable confidence in the measurements taken.

Adjust anything and you have changed the system, rendering the previous history highly suspect, if not useless.  In the days of trimpots, any adjustment changes the mechanical state of the contact between the wiper and resistive element where mechanical settling through pressure, vibration and/or thermal cycling would ensure you need to start a new history record and it would need a number of calibrations (comparisons to a standard) over a period of time to establish the pattern of change.  Certainly, the modern process of storing calibration data digitally reduces many such susceptibilities, but not everything.

[Fungus]

Where's the example of a meter that went out of spec?

https://www.eevblog.com/forum/testgear/hp-34401a-reads-a-little-low-on-ac/msg3241784/#msg3241784

Your own posts suggest that meter was somehow damaged (overvoltage?) and replacing some resistors put it back into spec with no other adjustment needed:

https://www.eevblog.com/forum/testgear/hp-34401a-reads-a-little-low-on-ac/msg3301010/#msg3301010

[bdunham7]

Your own posts suggest that meter was somehow damaged (overvoltage?) and replacing some resistors put it back into spec with no other adjustment needed:

Maybe that was the cause, I never got enough information to really know.  But the point is that I had an apparently nice working meter that was just a little out of spec.  Does the cause matter?  The point is that if accuracy is critical, just assuming that meters don't go out of spec, or even extrapolating assumed drift from previous observation, is insufficient. 

[Fungus]

Your own posts suggest that meter was somehow damaged (overvoltage?) and replacing some resistors put it back into spec with no other adjustment needed:

Maybe that was the cause, I never got enough information to really know.  But the point is that I had an apparently nice working meter that was just a little out of spec.  Does the cause matter?  The point is that if accuracy is critical, just assuming that meters don't go out of spec, or even extrapolating assumed drift from previous observation, is insufficient.

I'm not saying that meters don't ever go out of spec, it's the slow, constant accuracy drift that needs a yearly correction that I'm doubting.

There's a whole legion of Fluke handheld owners on here who are always proud to tell us that their meters are still bang-on after 20/30 years.

(and I believe them!)

There's also a few 1980s Radio Shack meter owners who say the same thing.

As for Anengs? They haven't been around that long yet but my money would be on them staying in spec even without laser-trimmed resistor arrays. We'll see... 

[alm]

Do DMMs and their components experience a constant drift? Yes. The most well-studied references are the calibration lab standards. For example, here is data for the 10V output of a Fluke 732A voltage standard from "PREDICTION OF THE OUTPUT VOLTAGE OF DC VOLTAGE STANDARDS", 2009, by Ilić et al:


And here is the 1.018V output, which is a resistive divider driven from the 10V output:


And here is data for the ESI SR104 standard resistor compiled by zlymex, that I grabbed from here:


It's even common do use linear regression to try to predict the value of the reference more accurately than assuming the last calibrated value. As described in the paper I cited. This would be a waste of time if the references did not have a constant drift.

A DMM will obviously have a less predictable drift, because there are more components affecting the accuracy, all with their own drift characteristics. Here is data from a HP 34401A calibrated three times in 12 years. I believe I got this data of the PMEL forum. Clearly the DMM is drifting, although based on this limited data it does not seem linear.


If the components in a handheld meter would drift less, the engineers designing these instruments would have done a horrible job . So clearly the handhelds will drift. Is this drift enough to be noticeable given their resolution? Maybe not.

Also, very high or low ranges will generally drift more, and ACV will drift more than DCV. I think the statements about the meter still being bang-on after decades is usually on something like a 9V battery, not a 10 MOhm resistor, or a 700V 20 kHz AC signal, which based on another report on 34401A meters that I pulled from PMEL was found to be out of tolerance 8 times out of 250 tests (the point with the highest number of OOT events):


Do you have any data on the meters being 'bang on'? Number of meters, range they were tested on, uncertainty of the test? How big is a 'bang'? Are we talking about agreeing to a least significant digit? The 24h tolerance? 1 year tolerance?

[bdunham7]

I'm not saying that meters don't ever go out of spec, it's the slow, constant accuracy drift that needs a yearly correction that I'm doubting.

There's a whole legion of Fluke handheld owners on here who are always proud to tell us that their meters are still bang-on after 20/30 years.

(and I believe them!)

Drift depends on many factors, but one big one, especially for heated references, is power-on time.  So a meter which is specified to drift less than a certain spec for a year might mean that they specify that the drift will be less than that amount even if the meter is on 24/7, and that with 95 or 99% confidence.  So the meter that is on for less than 100 hours per year probably won't experience that.  Yes, the vast majority of 3-4 digit DMMs will not go out of spec even over a very long time--and drift is probably less likely to be the issue at that level of accuracy..  Its not a conspiracy, it is just statistics.

[bdunham7]

If the components in a handheld meter would drift less, the engineers designing these instruments would have done a horrible job . So clearly the handhelds will drift. Is this drift enough to be noticeable given their resolution? Maybe not.

I'm not claiming that they don't drift, but if they aren't powered up much, drift may be the least important component of their actual uncertainties.

Also, very high or low ranges will generally drift more, and ACV will drift more than DCV. I think the statements about the meter still being bang-on after decades is usually on something like a 9V battery, not a 10 MOhm resistor, or a 700V 20 kHz AC signal, which based on another report on 34401A meters that I pulled from PMEL was found to be out of tolerance 8 times out of 250 tests (the point with the highest number of OOT events):

It's probably true that the commonly measured 10.00000 volts is the most accurate and stable range on the meter--as one might expect--but you need to include the uncertainties of the calibration equipment, especially when you look at those more difficult ranges.  The single 34401A example really shows this.  Also, your last example actually shows pretty much the expected result--a 96.8% rate of within-limits vs a specified 95% confidence interval.  So simple statistics would indicate that two thirds of the meters would actually be off by less than half of the specified, which for that signal is what I would call "bang on".

EDIT:  I noticed later that 700V @ 20kHz seems to be a nonstandard calibration point and I didn't get the same uncertainty from the 1-year specs.

Do you have any data on the meters being 'bang on'? Number of meters, range they were tested on, uncertainty of the test? How big is a 'bang'? Are we talking about agreeing to a least significant digit? The 24h tolerance? 1 year tolerance?

I have 8 Fluke 8840/8842A type 5.5 digit meters here in various conditions ranging from mint to war ravaged and each of them at least meets their 90 day spec on the 2 and 20V ranges and probably the 24-hour spec.  I haven't verified the other ranges yet but I haven't found any out of spec to the best of my ability to measure them.  None of the are anywhere near recently calibrated.  These are probably a bit exceptional in that regard due to the type of reference they use.