Fluke 5440A repair

[Rax]

Here's a 12hrs log (overnight) of the F5440A as read by the P6048. There's a garage door open in there somewhere (possibly where readings climb) - but very mild weather here, so I don't think that created wild temp differentials (also, electric car, so none of the heat a gas car would bring into a garage) - and a ferrite core clip-on the USB cable (possibly where the readings go down). I saw an effect of about .1-.15ppm descent as soon as I connected it, so there may be some truth to contamination over GPIB, but not positively conclusive (in no small part because I didn't really run that as a disciplined experiment, so that's on me).

Now, all that said, the stability doesn't look all that impressive to me. There's a differential of over half a ppm over half a day - is that reasonable/expected? I'm used to seeing about half that from this unit, though via "analog" observations, so obviously less reliable than this. I have not logged temperatures during the below readings, but again, I don't think they could have exceeded 1.5C, maybe 2C in the worst imaginable scenario.

[Rax]

Here's a roughly 36hrs log. A few notes:

• As I said before, the instrument hovers above 10.0000000V most of the time. It doesn't seem to be confused at all about where 10V is (Prema's, of course!... wink to the hardcore). It'd hover over that exact value for hours at a time - and by that I mean less than .1ppm off it (with occasional and typically brief excursions of about .15-.2ppm).
• There are a few hiccups. The first "peak" is .75ppm high... what could trigger such a jump? Anyone's guess, though certainly experience wins the big prize here. I have not seen these so distinctly before setting up automation... this was rather unexpected. The magic of 20/20 hindsight due to automation and logging.
• A couple other hiccups are at +.50ppm (just about the middle of it) and another one at about -.25ppm (at the very end). None of which I think are outrageous - my lab conditions are not "metrology grade..." And yet, the F5440A soldiers through those hostile conditions quite successfully, I'd say.

There's about 2.4C delta over these two days.

[alm]

Nice, so now you're getting actual data. It would be interesting how this looks over the longer term. It looks like there might be a slight negative trend in this data, but the period is too short to tell. This might be a temperature effect, which we'll know once you start logging temperature.

[Rax]

Here's continued readings - about 7k data points on this particular one, a little less than 72hrs 40hrs.

Keep in mind I haven't done an INT CAL (or a nulling for that matter) in a long time. Technically, the F5440A should undergo an INT CAL daily, though in case it's continuously powered, I think the advice of people having used this in cal labs was at least every two weeks I seem to remember.

[Rax]

Though most of the readings are below 10V, there's a constantly recurring crossing into the >10V side. The last 24 hrs are almost "dead on" (by which I mean eyeballing an averaging point hits 10.0000000V as precisely as one can tell). I'll try to average these to see what I get.

[Rax]

Averaging the ppm values, I get an overall .0135ppm, which include the peaks mentioned earlier (which I don't think are guaranteed to occur). In fact, the actual average without those is probably slightly negative, but smaller in absolute value.

[alm]

It would make the graph a lot easier to interpret if you include time on the X axis. It can be relaltive time, which could just be somehting like sample number * 20 seconds / (3600 seconds / hour) to have a scale in hours from t=0. Though tracking calendar time is useful if you want to relate it to other events like temperature measurements or daylight / night.

I don't think averaging over such a long time scale is useful. My reasoning is that on such a long time scale, the result will be a combination of tempco, time drift, etc. So either you calculate the average over a period of let's say a couple of hours max to estimate the value at a single point in time (though likely averaging a couple of dozen samples will give you pretty much the same information), or you do some sort of regression analysis to measure the influences of external factors like time and temperature. This will be difficult on such a short time scale since time and temperature differences will be very small, but the longer you track the easier it will be to see particularly drift over time. What could help spot trends is applying a rolling average filter to filter out some of the noise.

If you report averages, then I think it's also useful to report standard deviation and number of samples in addition to the average.

[Rax]

Nice, so now you're getting actual data.

Thank you, alm! It's been a steep couple of weeks-ish, but with the help of good, knowledgeable, and generous people here I got that running. I still can't really believe it (that's how much I enjoy it).

[Rax]

For shits and giggles, here's the flattest (and closest to Prema's 10V) ~22hrs (about 4k data points) stretch from those 72hrs. Yes, it absolutely is cherry picked data, but I think it's both satisfying and still useful information to ponder over.

Not sure if temperature, noise/EMI/other environmental garbage, or other conditions determine the peaks and valleys on the overall run, but I see this clean stretch as a statement of the fine quality of the F5440A.

_______________________________________________________________
Added time in hours on X axis for clarity and legibility.

[Rax]

If you report averages, then I think it's also useful to report standard deviation and number of samples in addition to the average.

I calculated the basic average using excel tools (AVERAGE, outputting .0135ppm), then standard deviation based upon the same (STDEV.P) comes up with .135ppm (assuming I used correctly the excel calculator, which I merely dabble in). The number of samples is in the original post (about 7000).

If I calculate the same on the 22hrs sample included above (about 4000 entries), I come up with -.0694ppm vs. .0719ppm, respectively.

I just kept on logging, currently. I'm interested in seeing how often these jolts tend to happen, and if there's any pattern or rule to it.

[Rax]

Continued monitoring this past the 60hrs milestone. I suspect there's some "real time" gaps on this, possibly related to saving into the txt file over the network (?). So even if we're at  72hrs into this, there's 11,040 entries into the enclosed graph, which amounts to a bit over 60hrs. I don't think that truncates the data very much (though it may account for some of the "jolts").

For this:

• The number of samples = 11,040
• Averaging = -.0038ppm
• Standard Deviation = .1239ppm

I'll stop posting here at some point, but I thought a bit longer term data would be interesting to present.

[Rax]

What could help spot trends is applying a rolling average filter to filter out some of the noise.

I don't know what you mean with this. Is this a data handling tool, something I can possibly do with my spreadsheet program?

[Rax]

Continued monitoring this past the 60hrs milestone.

BTW, goes without saying, some of those jolts could as well me "thermic shocks" by opening the garage door, etc. This should be at least twice a day, so it looks like most of the time it had virtually no impact on the measurements whatsoever (I regard anything under, say, a .2ppm fluctuation as essentially noise).

[alm]

I don't know what you mean with this. Is this a data handling tool, something I can possibly do with my spreadsheet program?

Yes, a rolling average or moving average is effectively a low-pass filter for data. It averages the last N points (where N determines the degree of filtering) with the goal of uncovering patterns that were previously obscured due to not so relevant high frequency. It should be trivial to find a tutorial how to do that in a spreadsheet, MATLAB/Octave, Python or whatever data analysis tool you use. I would use it as a plot in addition to the raw data plot.

[Rax]

Still learning my instruments. And probably fundamentals on all of this.

After the perfect agreement (which didn't seem to go away for about six days now of continuous running, while leaving the instruments alone, and I kept tracking logs to document it past what I already posted here), I decided to experiment some more, and leaving the P6048 on (and not touching it), run a fresh INT CAL on the F5440A. The result is essentially the same 1.5ppm lift-up of the reading on the P6048. The symmetry of this is, probably, relevant - for reference, I refer here to the identical event where a nulling of the P6048, while leaving the F5440A alone, had apparently the exact same effect - and, surely, has a certain beauty in the spirit of ancient Greek esthetics. Which, for the record, I'm not a big fan of (symmetry as beauty).

I am reviewing several prior pages to this thread to see if anything in the feedback gives me a basis to work with to explain this (other than that anything around 1ppm and lower is subject to fairly random laws, which I regard as legitimate). Being symmetrical, I'm thinking:

• EMF potential over my cables (low-emf AB sourced). Though I think dietert1's input was emf shouldn't account for this much.
  
• Some sort of consistent tempco (can't imagine how).

From the possible hypotheses presented this far, nothing stands out to me as a compelling explanation. I know we've circled this issue quite a few times already, but after performing the parts asked to perform I think these jumps still remain unsolved. Or at least to me, but if I am missing it I'll appreciate a pointer. Possibly missing the forest because of the trees.

It looks like any of the long term logs I am recommended to do rely on explaining this persistent realignment (or else I risk random, but persistent 1.5ppm offsets). I am trying to recall how this went away the last time, but I think it may just be nothing other than time passing. By which I mean given enough time the two instruments seem to fall back into agreement regardless what operations (nulling, INT CAL, etc.) have been  performed in the meantime.

[Rax]

I should also add that my prior observations of INT CAL (F5440A) + nulling (P6048) would have typically generated convergence, not divergence. Hence, as I'm observing this new (?) consistency of the divergence (and of course, these few observations don't amount to a statistically relevant body of work), I am trying to decrypt the patterns.

If I'm overthinking this, please call me on it.

[Rax]

(after doing some homework, I edited my prior two posts - hopefully that doesn't make this confusing)

[Rax]

As I'm currently having that 3458A/002 I mentioned here and elsewhere before, here are the SM performance tests done with it. This is a unit calibrated about two years ago at KS' Rosemont facility.

I include a snip of the allowed ranges for compliance.

• 1.10uV
• 99.99616mV
• .99999535V
• -2.70uV
• 1.00000220V
• 5.0000036V
• 20.000001V
• -19.999998
• 99.999184V

I didn't do the 1000V, as I'm a little antsy with that with with SELV cables and wnatnot.

This looks good to me.

[RandallMcRee]

As I'm currently having that 3458A/002 I mentioned here and elsewhere before, here are the SM performance tests done with it. This is a unit calibrated about two years ago at KS' Rosemont facility.

I include a snip of the allowed ranges for compliance.

• 1.10uV
• 99.99616mV
• .99999535V
• -2.70uV
• 1.00000220V
• 5.0000036V
• 20.000001V
• -19.999998
• 99.999184V

I didn't do the 1000V, as I'm a little antsy with that with with SELV cables and wnatnot.

This looks good to me.

What I expected: comparison of 3458A with cardinal points of 5440. E.g. for 5.0V entered on 5440 3458A reads 5.xxxx volts....for 10-20 such points.
What I see: a table from somewhere or other having nothing to do with this?!

Am I being dense? Sorry if so!

[dietert1]

I 'd guess the first table are the test measurements and the second table are the specs. It's not so obvious whether the 20 V and -20 V results are for polarity reversal of the calibrator setting or for the meter input cables. At least i understand that the measured deviations at 20 V are below 1 % of specs and much below 1 ppm, a very satisfactory agreement. A christmas present if you want so.

Regards,

[Rax]

I 'd guess the first table are the test measurements and the second table are the specs.

That's right. I was closely following the performance verification procedure from the SM.

It's not so obvious whether the 20 V and -20 V results are for polarity reversal of the calibrator setting or for the meter input cables.

The procedure doesn't clarify that, but I used the polarity reversal control on the 5440A.

[bdunham7]

What I expected: comparison of 3458A with cardinal points of 5440. E.g. for 5.0V entered on 5440 3458A reads 5.xxxx volts....for 10-20 such points.
What I see: a table from somewhere or other having nothing to do with this?!

Am I being dense? Sorry if so!

Back when these were new it was not easy to read those voltages to that precision, so they set up the performance test and calibration procedures to use the least number of tests that could verify proper operation provided the instrument didn't have some large or weird malfunction.  Given that there is a 3458A present, I think doing a many-point linearity check would be the way to go.

Also, I think the apparent 40ppm tolerance for the 1V setting is a typo.

[Rax]

The procedure was performed at 24.9C, in case anyone was wondering. The 5440A got a fresh INT CAL and the 3458A had an ACAL the same day.

[Rax]

Also, I think the apparent 40ppm tolerance for the 1V setting is a typo.

In the manual, you mean? That doesn't make a whole lot of sense to me either.

[Kleinstein]

I 'd guess the first table are the test measurements and the second table are the specs.

That's right. I was closely following the performance verification procedure from the SM.

It's not so obvious whether the 20 V and -20 V results are for polarity reversal of the calibrator setting or for the meter input cables.

The procedure doesn't clarify that, but I used the polarity reversal control on the 5440A.

As the specs / test is for the calibrator, it should be the polarity reversal in the meter. For better accuray of the measurement one could consider also swapping the meter leads, so that the meter reads the same sign in both cases.  20 V is not so much a sweet spot for the 3458 and the meters turn over error (e.g. offset in the 100 V range) may be about that large.
With the 3458 I would add a +-10 V test to see if the turn over works Ok. I would not expect much difference as this is only a polarity relay - so mainly a little offset from thermal EMF.