Improving Cumulative Uncertainty For Multiple DC Reference Standards

[Hawaii596]

If anyone here has extensive background with things such as Fluke 732A, 732B, etc.,  here's a question I'd like some thoughts on:

I recently inherited a number of DC References (multiple Fluke 732A's, 732B, and a Datron 4911 (four references and an average output).

I'm plotting them now to see how they will trend compared with each other and maybe pull the worst and use the ones with the best trend.  I haven't sent anything out yet (or had a unit sent in for a "hot cal"), as this is still in the development stages.  I also have to tighten up the room temp, etc (as I'll also be bringing my Guildline 9975 on line some time down the road).  And, of course, my Fluke 792A, etc..

So, back to the DC references....  I want to develop a statistical model to somehow incorporate the average of multiple outputs to either optimize uncertainty numbers or at least optimize confidence in the existing uncertainty numbers.  I'm thinking I'll need at least a year's data before I can incorporate multiple combined uncertainties, so I'm going to do a variety of short and long term plots on them and analyze relative stabilities (then incorporate an absolute number a little later when I get one or more of them cal'd (in prep for "going live" with them).

Are there any accepted statistical models for cumulative uncertainty when using multiple DC references for main DC traceability?

[Bryan]

I don't claim to be an expert but I think you may need a little more time unless you have some good historic data for the things.  I have a couple 732As and by setting up a spreadsheet to record reported values (from our support agency), time between reporting measurements I was able establish a pretty solid drift rate/day, I apply this my last reported value and believe it serves me well.  I have projected what the 10V reading will be to some of my coworkers when I send it for cal and am typically within 2 microvolts, guess that is 0.2ppm at 10V.
I'm kind of riding the edge though because I only have 2 that I really trust, you should have at least 3.  That way if one heads south you can figure out which one it is.
I use an 8508A and do a transfer measurement, that is if I know mine is 10.xxx volts and the 8508A reads 10.yyy, the one I am measuring can be calculated by a cross multiplication.  An auditor suggested I try a differentail measurement on 2 732s at the same time, it was simpler & the results were comparable.
There is a paper from Fluke, an application note titled "Fractional ppm traceability using you Fluke 734A/732B".  You can get it from their site in the support area, requires registration.  I think you will find it interesting.
Good luck

[Hawaii596]

Oooo...  Good find.  I'll look that Fluke paper up.  I have 4 - 732A's, 1 - 732B, and 1 - Datron 4911 (which has four cells, and an "AVERAGE" output as well).  Not including the "AVERAGE" output on the 4911, I have 9 each 10V References.  The 1 year spec on the 10V AVERAGE output on the Datron 4911 is 1.0 ppm.  For anyone's curiosity, I owned two, inherited a couple more a couple of years ago, and just inherited the rest recently from closed facilities.

The other question will be about how to minimize cal costs.  If I have a "HOT CAL" done, maybe get my best reference done in an annual hot cal (maybe quarterly for the first year while I'm gathering data); and use the HOT CAL unit and the associated data in statistical combination with my ongoing readings of all the references to establish values for the rest.

Regarding plotting the deltas,  that is currently in my "equation" (so to speak) as for how I am plotting.  Since intercomparisons are relative, as I'm establishing which are my good and not-so-good references, I'll be looking for which units give me my worst outliers.

I do have a couple of 8508A's with the good Fluke cal on them.  But unfortunately, for now at least, I am using a 3458A.  I did take my 732B out and adjust it to the certified 10V numbers on an 8508A (remembering this whole array is not yet in "legal cal" - still in development).  So I'm using that as my starting point.  So I'm "relatively" within about 0.1 ppm between those two. 

I'm going to take daily readings for a while, to get some relative short term drift numbers, then go to weekly (probably about amonth out).  I'm presuming this will take about a year to amass sufficient data for a good starting point. 

The other point I need to factor in is temperature.  I should probably place a good RTD right next to them and plot and account for temp coefficient to see how much they are each impacted (perhaps even try to calculate actual temp coefficients on each).

And finally, I notice a little variation between them in instantaneous variation.  If I use the 5440A cables, guarded at the DMM and reference, LFILTER ON, NPLC >100, NDIG = 8, and monitor lets say, 10 iterations on the DMM, the max and min measured values on the DMM (after I allow a few for stabilization) seems to vary a little between references.  I' thinking I should do about a 100 reading experiment and plot Std Dev on them to see if there is any significant difference, evaluate as to whether it is a consideration, and use that as a part of my criteria for which one will be my most golden unit(s).

I'm also thinking I should do battery testing early on and replace batteries as soon as possible, so I'll have a fully robust set of standards throughout the measurement period.

There's my morning ramble. 

[scottbp]

I have a Datron 4910 "four-banger" 4 output zener standard as well. The difference in the 4910 and 4911 is that the 4910 has a 4-wire output buffer and two divided outputs, one for 1.000 volts and one for 1.018 volts. The 4-wire buffer is so you can connect it to a Fluke 720A or 752A voltage divider (if you connect one of the cell outputs directly to a 752A, it loads it down-- ask me how I know).

Anyways, I use a Dataproof scanner and Dataproof Voltref software to track the cells, and plot each individual cell and the output on a linear regression chart. Any time I need to calibrate our 5720A or 3458A, I run an intercomparison to get a "daily value" or I look at the chart I've been keeping (if it has been running stable). I use a Fluke 8508A to do the intercomparison, but it allows for a 3458A, a Datron 1281, a HP 34420A, and I've even got good results with a Solartron 7081. Voltref does have a charting function built in, but I've been keeping it on an external spreadsheet.

Whatever you do, don't ship the standard cells!  Especially not via the big brown elephant (you know which one that is!) Hand carry them to your nearest lab that has a Josephson-Junction primary DC voltage standard (you know, the one that's kept in the big keg of liquid helium), be it a NASA lab, a national lab like Sandia, or wherever. For me, the nearest is the Lockheed-Martin Integrated Metrology facility at the NASA Stennis Space Center in southern Mississippi. I then compare the results with my chart, and for the last four years running it has been well within the predicted uncertainty.

On a couple of occasions, some of our customers have sent in their Fluke 732B standard cells. I first intercompare it with mine, then send it to the Lockheed-Martin lab, then intercompare it with mine again for an extra vote of confidence, as a round-robin of sorts. And the best part is- the customer pays for it.

[Hawaii596]

I believe mine is a 4911, as I only have 10V and not 4-Wire outputs.  For an "off-brand" (not so far off), I was pretty surprised.  I guess the Datron is a pretty good unit (I never cared much for some of their other products).  This is not an "IN CALIBRATION" activity yet (don't want to spend cal funds for this until I get everything up and running), so I'm using best available checks.  I used the 10V numbers from an 8508A (that I can't tie up in that area as it is used for real cals), and the 4911 which hasn't been cal'd since 06 was within less than 1 ppm. 

I'm not sure if anyone here in the local area has a Josephson junction.  I'll have to make some calls around to find out. 

As these are all coming out of mothballs (some of them have remained continuously plugged in), I take you would concur that right at the start I should make sure I have good batteries on all of them?

ONE MORE QUESTION:  I just downloaded the Fluke white paper and started reading it.  They mentioned NBS(NIST) Tech Note 1239.  Do you  (or anyone else) happen to have a copy or know where I can get one?  OR, is it worth the read?

[purpmcrider]

I might be able to help you out with some of the details you are looking for and have some insight into the differences you are seeing with the 732A and 732B.  I recommend down loading every paper you can get from the Fluke website that Les Huntley or Ray Kletke wrote on the maintenance and use of these standards.

If there's not enough there, send me an e-mail. . .  I might have some papers that are no longer posted.  Are you measuring the outputs of the standards by means of a transfer or are you putting the units in series opposition? I didn't see that spelled out. . . sorry if I missed that.  The 3458A should be very capable of supporting the measurments you are trying to make.  Fluke used an 8505A until 2005 to support the DVMP in the Primary Lab.

Let me know if I can help.