288 hours isn't long term. Temperature drift is still drift, so it ultimately counts towards the accuracy of the system as a whole. This is why those figures are given in ppm/C. For long term temperature-independent results, the unit needs to be placed in a small chamber that can maintain absolute temperature (25c, 40c, whatever). Lastly, if you've only got 6.5 digit meters, don't bother. At that voltage the 10uV decade is noise anyway and accuracy to 100uV is nothing special. (Power Designs was doing 10uV accuracy in their benchtop power supplies in 1964 with a Zener and a few resistors).
Sent from my Tablet
I'm just reporting what I have so far. If you don't like it that's really not my doing. But I'm impressed with the performance thus far. You should be too. As I said the test is "and running". I can't speed up the passage of time. But you can still criticize me for not doing that. I don't know what drives you. But those who contradict everything another man says, and who criticize everything another man does, are the sand in the intricate mechanism of civilization that makes cooperation among men impossible.
I like how you automatically go on the offensive without reading (or understanding) my entire post.
Let me break down what you need to do for true long term testing:
1) You need to test at least 10 units at the same time. This can easily be done for less than $100 by buying a switching mainframe on eBay. In fact, there are a few HP 3488A up right now for >$75 with relay modules. Another good choice would be the Keithley 7001, it runs a bit high price wise but you can find them for less than $100. Finally, the Keithley 705 is a solid choice. (If you really wanted to do it right you would get a Keithley 2002 8.5 digit meter, which has a built-in slot for scanner cards.)
2) All 10 units need to be in a controlled, constant temperature environment. A styrofoam cooler with a 40c PTC heater would work if you don't want to deal with PID controllers.
3) The wires going to the scanner need to be shielded twisted pair. The wires going to the DMM as well should be terminated with gold copper spade lugs for low thermal emf affects.
4) You would have the scanner switch between units every X seconds and trigger the meter to make X measurements at the slowest integration rate, for each unit, which your software will then average and store as a single reading.
Do this for 3-6 months undisturbed and you'll have useful data.
In fact, I couldn't tell from your last post, but you *are* logging what you're currently doing to a PC and not relying solely on the compressed averaging TrendPlot display, right?
Sent from my Tablet
That's a good design for the test. But it isn't the only possible design. It's a brute-force design needed to determine a minimum error. The other way it can be done is to use the worst-possible-scenario example, measure the results, and determine the maximum error of the worst possible example. Other units will be equal or better. Your design is meant to produce data to support tighter specifications. The alternative to that is to spec the device with comfortable margins for error. Then, you just have to determine whether any example performs outside the margins. Setting the error at double what experimental evidence indicates saves a lot of work.
We have an 8-1/2 digit meter. I only need it for checking the results. I'll confirm the findings using 8-1/2 digits. I use coax interconnects. I know that seems a bit weird for DC, but it eliminates several unknowns. The environmental chambers use PWM controllers, which create a 15A DC pulse at 150MHz. Twisted pair is susceptible to induction by high-frequency, high-current pulse signals nearby, but coax + BNC is not; at least not perceptibly so.
If we decide to try to squeeze another ppm into the accuracy of this design, then I will use the test as designed by you. But right now we're just establishing absolute boundaries. After 14 days, the long-term drift is still masked by temperature drift. It would be nice to know "typical" performance. But for now we're using worst-case figures.
Accuracy figures for instruments are based on worst-possible figures, given the most hostile environments, the most external interference, and the worst possible drift the instrument is rated to endure. In industry, accuracy of measurements in a system must be assured, or it could make the difference between producing a useful drug or a deadly poison. In a controlled lab environment, an instrument should remain consistent across calibration periods.
The absolute error ratings apply at temperature, radiation, vibration, pressure and humidity extremes. I've had meters that remained well within spec for 10 years, or more. It is proper to respect the volt; to practice due diligence and competency in procedures, methods, practices, documentation, etc.. But it is also proper to use rational sense: if a meter has drifted 0.5ppm during the last decade, and it is used in a laboratory, the maximum error figures given by the manufacturer apply only to an auditor who uses those figures as a guide to determine whether or not the device requires repair or replacement.
The instruments I use are optimized. Calibration and optimization are different things. Most calibration involves no adjustment. The device is simply checked to determine if it's within spec. If you want a piece of equipment spot-on, you have to pay for optimization, or adjust it yourself.
In real life, if you've got historical data, and the device is reliable, it probably hasn't drifted much since the last calibration. If it has drifted so much that it is out of spec, someone would notice, unless it's the only meter in the joint.
I appreciate the equipment recommendations. I save money when I can, which isn't often. You're correct. If I want to obtain the most useful data, it's going to take time and effort. Data puts more value into the product, but practically no one actually requires +-2ppm. It's nice to have. But I think I'm maxed out on the price as it is. I sell more +-6ppm units at 59.95USD than I do +-2ppm units at 97.95USD
I appreciate your suggestions and I learned a few new things. Thank you. I'll check out the equipment you recommended. How are the switching units rated, what spec do I look for?