[pic heavy !] Mineral oil stabelized voltage reference

[Spikee]

Recently i have acquired a Fluke 87 series III and the idea of calibration came up.
As a hobbyist it is most likely not viable to get it calibrated at an company.

So i was thinking of making my own voltage reference. You want a reference that has a low temperature coefficient but also a good initial accuracy so you can use the calibration transfer method. But finding good initial accuracy and low temp drift can get expensive .

The old 8 1/2 digit multimeters use some LT reference that has an accuracy of like 4% and 0.05 ppm  but they cost ~40 USD and because of the inaccuracy of it you can't do the transfer method without having a calibrated 8 1/2 or 6 1/2 digit dmm.

mineral oil idea scrapped: Because of the good feedback i learned that it is not a good idea and that there are better methods of dooing it.
Url of the new "plans" : https://www.eevblog.com/forum/projects/mineral-oil-stabelized-voltage-reference/msg178100/#msg178100

Code: [Select]

To use the transfer standard method you need a good initial accuracy like the LTC6655 (5v ref) 0.025% with a temp co of 1ppm.
And to solve the temp co problem just put it in an temperature controlled environment.

I was thinking about putting it into a water / oil proof metal enclosure:

[img]http://www.hammondmfg.com/jpeg2/1590Z120_OB1.jpg[/img]

fill it up with (non conductive) mineral oil :
[img]http://www.ikea.com/be/nl/images/products/skydd-houtolie-voor-binnen__15140_PE099029_S4.jpg[/img]

put a cheap 220V 100W heating element in it from china @ 3.99$ :
[img]http://m2.uxcell.com/i/12a/ux_a12021000ux0260_ux_c.jpg[/img]

and buy the cheapest pid controller on ebay 13.99 usd:
[img]http://i01.i.aliimg.com/wsphoto/v1/581386509_1/100-New-Dual-Digital-F-C-PID-Temperature-Controller-REX-C100-PID-Thermocouple-0-to-400C.jpg[/img]

The cheapest screw k type thermocouple you can find @ ebay for 1.67 $ :
[img]http://www.sz-wholesaler.com/userimg/505/546sw1/flexible-thermocouple-screw-type-659.jpg[/img]

Of Course  use the proper IP68 cable glands so your cables can run from inside the enclosure to the outside world:
[img]http://edgecdn.lappgroup.com/typo3temp/GB/74d85f0dc4.jpg[/img]

[list]
[li]solid state relay to switch the heating element on / off 5$[/li]
[li]all the other stuff i forgot ...[/li]
[/list]

It should be filled completely with oil so no condensation can form inside the case.
The temperature would we regulated to like 50.0 degrees Celsius . The LTC6655 has an feedback connection so it will compensate for the voltage loss in the wires so you can use kelvin connection. The LTC6655 is around 8 USD .

The low temp co and high accuracy vs price point is especially noticeable for resistors. Good ones are around 20-40 USD for one resistor ...

This all seems like a nice experiment / build . The month of January is very busy for me so it will have to wait till next month before i can do some proper testing.

stuff i already have laying around:

[list]
[li]Solid state relay[/li]
[li]PID controller[/li]
[li]Screw K-type thermocouple[/li]
[li]Loctite 5926 silicone gasket[/li]
[/list]

Any ideas / feedback will be really appreciated !


my lab equipment:
Agilent U8001A 30V 3A linear lab supply (in cal)
Delta elektronica fixed 5V 10A linear psu (out of cal)
Rigol Dm3058 5 1/2 DMM (out of calibration 1 1/2 years ) (via pc you can read the full 6 1/2 digits)
Cheap voltcraft current clamp (out of cal)
Fluke 87 III dmm (out of cal)

[Jay_Diddy_B]

Hi,

I would consider not using an oil bath for the following reasons:

1) Around normal room temperature the tempco is flat:



2) Oil baths used in calibration labs are 'stirred' to minimize temperature gradients.

3) Consider the drift during the first 1000 Hours:



This comes from mechanical stress on the silicon relaxing. The ceramic package LS8 version has much better performance:



This only comes in a 2.5V version.

Jay_Diddy_B

[TorqueRanger]

I don't know about everyone else but this just seems like a lot of work just for calibrations...

[SeanB]

100W is way too much heating, If you insulate the box well ( 30mm of polystyrene foam on all sides) a 10w heater will keep it at 70C with no problem. I ovenised a crystal years ago in a housing made from cardboard and thin 1mm polystyrene foam sheet, and the heater there is a TO220 transistor dissipating 200mW, which is enough to keep it stable.

You will need to lead all openings out of the lid, and those glands sadly will not do, you need smaller glands ideally a multipole connector with an IP68 rating and a gasket compatible with the oil that you use to connect to the box. As well you need to have a filling valve and an expansion port, preferably with a small reservoir to hold oil as it expands with temperature and external pressure. Your oil needs to be ultra pure, dry and degassed, along with having no conductive additives. Best bet will be a mineral refrigerant oil, as it is designed to be low foaming, low water and work with motors. Do not use a POE or PAG oil there though, they are not going to work well, one is conductive (PAG) and the other is incredibly hygroscopic. Also note that all wiring inside will have to be made in either PTFE or oil compatible insulation, so no PVC or silicone, only fibre insulated or PTFE coated wire.

Better is to insulate the outside and bolt a 10W resistor to the box, and provide internally a smaller metal ( made from PCB blank) box that holds the reference and associated support and which is placed on standoffs to keep it in the middle of the outer box.  An outer ventilated housing with the controllers and such will contain it all then. Then the wiring can go through a set of drilled holes with sleeving and it will be quite stable.

[Jay_Diddy_B]

Hi,

I had another look at the temperature graphs.

Between the 25 and 50 degrees C the typical change is 2mV this is 0.04%.

This is similar to the specifications of the Fluke 87

The Fluke 87 is 4 1/2 digits 20000 counts

So the temperature coefficient shows up in the last digit as +/- 2 counts

4.998, 4.999, 5.000, 5.001 or 5.002

Which is fine for a quick check.

I would build the circuit. See if you are satisfied with the performance, before proceeding with the bath.

Remember the reference in the fluke DMM is not in a controlled environment.

You can also look for a Fluke 731A. These sell on eBay for around $140.00


Jay_Diddy_B

[mikeselectricstuff]

Oil would get real messy  - what about thermal expansion? Normal IP68 deals may not be specced to be oil resistant.
a PCB with a groundplane, thermally insulated should stay pretty much isothermal. If necessary screw it to a block of aluminium.

[ftransform]

idea though. Why bother using all that expensive/difficult to do IPA underwater stuff and leave the connectors on top?

unless you are into building milspec shit I think you could do better by using a hole on top + hot-snot or silicone sealant (you would have to cut it if modified tho).
So long you take reasonable care to keep it upright you could use a Tupperware container or a small water bottle. Also maybe Styrofoam?

I feel like getting all those o rings and what not in reasonable quality is pricey. Having a cup of oil leak in your lab would suck too (imagine if it dripped onto your test gear!), so keep it in a tupperware anyway whatever you decide to build. Minerals not a bad choice.

[Fsck]

Also, without knowing the error of the thermocouple and the controller, you just screwed your voltage reference in accuracy. k-type thermocouples aren't known for their accuracy.

[Things]

K-type thermocouples can be horribly inaccurate indeed! I've got one here that's reading a whopping 12C over ambient room temp, brand new too.

[cwalex]

I know this goes agains the diy approach but you can get a dmm check from voltagestandard.com for about $50 I think. They are calibrated and can be sent back for recal for about $5 bucks or something.

[Spikee]

i guess i just should try it.
I already ordered a dmm check and it should arrive soon.
The LTC6655 is ordered and i will see if i can thermally insulate it without using mineral oil.
It is also possible to just put the reference in an oil bath and leave all the heating stuff etc out of it. In that way all the components will be at the same temperature after some time. I already looked for a dc standard locally but i cant find one for a reasonably price.

Will this design suffice ? :
http://video.linear.com/28

i got samples for those chips a few months ago

[mazurov]

Slightly OT - does anyone know the time drift figures for good DMMs? For example, I have 6.5 digit HP 34401A which will have its calibration expired in 2 weeks. I don't really care about this kind of accuracy - shall I worry or can I just treat it as 5.5 digit DMM for the next 3-5-10 years?

[SeanB]

Depends on environment it is kept and used in. Temperature stable environment and it will stay within cal values for decades, normally they do not drift unless they are moved ( vibration) or have large temperature swings. 30 years and still within cal values is common for these style of high accuracy references.

[e100]

Put the voltage ref together with a 1W heater and temperature sensor inside a small section of copper water pipe filled with heat sink thermal grease and put the whole lot inside a vacuum flask with long thin wires that exit through the top.  It may take several hours for the PID controller to self tune and for the interior mass to reach temperature equilibrium. Use bare copper type connectors for the voltage output to avoid creating unwanted thermocouple voltages.

[Gall]

I would suggest using a copper block with drilled holes for a small resistor for heating and for the sensor. Thermal insulation may be made out of cotton or polystyrol foam. This needs only about 50-100 mw for heating.

Anyway, THE best method of getting precise temperatures is using either melting ice or boiling water. Melting tin is commonly used for thermocouple calibration. Other chemicals may be used for getting other temperatures.

Also there are many physical effects that do not depend on temperature much. My personal favorite is the technique invented by one of my father's collaborants. He stabilized voltage of about 50 kV with less than 1 mV precision by closed-loop locking at the intensity of characteristic x-ray emitted at that voltage.

[Jay_Diddy_B]

Hi,

If you look at these curves:



There are two good place to operate the reference one is around 35C and the other is around 90C. At these two temperatures the reference has the least sensitivity to temperature.
Of the two, the lower temperature is probably better. Operating at the higher temperature will have issues with thermal hysteresis.

Jay_Diddy_B

[Spikee]

I should do some more research about that.

[mzzj]

Recently i have acquired a Fluke 87 series III and the idea of calibration came up.

Forget Chinese temperature controllers and k-type sensors if you want any sort of stability.  Thermistor or Pt100/Pt1000 sensor in a simple control loop set to fixed temperature does much better.
You don't actually need much of  (absolute) accuracy in your temperature control, just stability/precision.

As already mentioned, LTC6655 stability looks pretty bad. Old industry  workhorse LM399 should be quite nice, but you need to calibrate it against some other reference. I would vote for voltagestandard.com varieties unless you have possibility to calibrate your homemade "standard" somewhere.

[Conrad Hoffman]

I wouldn't bother with the whole temp control issue. Years ago I wrote an article on building a voltage standard-
http://www.conradhoffman.com/mini_metro_lab.html It's a bit long of tooth, but the concepts still apply today.

Though you can start off with a close tolerance part, IMO it's better to start off with a very stable part, age it a bit, then just measure it on somebodies recently calibrated 6 1/2 digit meter.

[Spikee]

http://www.conradhoffman.com/mini_metro_lab.html It's a bit long of tooth, but the concepts still apply today.

I saw your article some time ago and it got me interested in kelvin varley dividers.
After some more research on the web i started searching for a 11 positions 2 deck switch .
At first it took some time before i understand ed that "normal" rotary switches have a index of 30 degrees (didn't know this ...) and because of that i wasted a few hours.

i found these switches on ebay: http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=300562748033&
(Amspire bought the same ones for his kelvin varley divider , but was sick so he didn't finish his build ).

I will make a seven decade kelvin varley divider . just because ... you can .
I still have to do some research about the resistor values that i will use , i also could just use the values of kvd-700:


I also read that the resistors need some burn in time so i should put them in parallel and connect them to my lab power supply and let it run for one or two weeks. After that i need to use the Wheatstone bridge method to select resistors that are closest together.

I will most likley buy a calibrated voltage reference from voltagestandard or geller labs because they are calibrated and can be sent back to be re-calibrated for a few dollar .

I searched a bit on the internet but i didn't find any diy kelvin varley builds so it probably will be hard.

I already ordered these parts:
7x 11postion 2 deck switch: http://www.ebay.com/itm/300562748033?
10x cool knob for the switch: http://www.ebay.com/itm/290605273560?

[Conrad Hoffman]

The trick with KVDs is finding good enough resistors to build the top few dividers. That's why you rarely see them with more than 6 or 7 decades. I think the Fluke 720 put adjustments on the resistors, knowing they'd never be good enough without them. I used regular 1% metal films that I selected, but the limitation was soldering them in. Every time you do a solder joint they have a tendency to move a few PPM. Once I got a chain of resistors that were correct, they were fairly stable.

It's worth trying to find some old Julie Research application notes. They made their resistors by winding wire on a multi section plastic bobbin, first one way, then the other, to cancel out inductance. That doesn't work very well because the coupling is bad but's better than nothing. Their real trick was using a very stable wire like Rediohm 800 or similar (can't be soldered, has to be welded) and doing the windings loose.  Tight windings stress the wire and change with temperature. Loose windings are way better. I've made lots of precision resistors, but only for low values. It's much harder to make high values with very long lengths of very fine wire!

On the voltage source, and standard resistors for that matter, the modern thinking is not to worry about adjusting anything to a specific value. It just causes stability problems and wrecks the history. Just make every fixed and get a calibration. Every time you have it calibrated, you add to the history. History is what gives us confidence!

[ftransform]

The trick with KVDs is finding good enough resistors to build the top few dividers. That's why you rarely see them with more than 6 or 7 decades. I think the Fluke 720 put adjustments on the resistors, knowing they'd never be good enough without them. I used regular 1% metal films that I selected, but the limitation was soldering them in. Every time you do a solder joint they have a tendency to move a few PPM. Once I got a chain of resistors that were correct, they were fairly stable.

It's worth trying to find some old Julie Research application notes. They made their resistors by winding wire on a multi section plastic bobbin, first one way, then the other, to cancel out inductance. That doesn't work very well because the coupling is bad but's better than nothing. Their real trick was using a very stable wire like Rediohm 800 or similar (can't be soldered, has to be welded) and doing the windings loose.  Tight windings stress the wire and change with temperature. Loose windings are way better. I've made lots of precision resistors, but only for low values. It's much harder to make high values with very long lengths of very fine wire!

On the voltage source, and standard resistors for that matter, the modern thinking is not to worry about adjusting anything to a specific value. It just causes stability problems and wrecks the history. Just make every fixed and get a calibration. Every time you have it calibrated, you add to the history. History is what gives us confidence!

How do they weld wire like that? Special spot welding device? Tig?

[Spikee]

I could try to put the switches whit the resistors soldered on them in an oven for a day or so at 40-50 deg. C. So they settle back to the same value. Would that work / have an effect ? I just received my dmm check so i am able to verify resistive values at 5 1/2 or even 6 1/2 digits. Ill have to carefully select the resistors that i will buy to get the lowest drift and temp coëfficiënt without spending to much

[Conrad Hoffman]

Predictions are tough, especially when they concern the future. Or resistors. Soldering temps are pretty hot, so try heat sink clips and an oven bake is always a good idea.

The spot welding is done with small capacitive or inductive discharge welders. Look up Ewald or thermocouple welders. They tend to be pricey but you can DIY one for next to nothing. The "800" series wire has a lower tempco than manganin if you can get over the welding hurdle.

[Spikee]

If i put the resistors in the fridge and put a heatsink on the leads will that. Be an inprovement or just the opposite effect

[mazurov]

If you have a good ohmmeter, just set up an experiment. Measure the resistance, write it down, solder, measure again, compare. Then repeat with another resistor, this time holding it with needle nose pliers between the body and the solder joint. Compare the difference in resistance induced by first and second technique.

[Spikee]

If you have a good ohmmeter, just set up an experiment. Measure the resistance, write it down, solder, measure again, compare. Then repeat with another resistor, this time holding it with needle nose pliers between the body and the solder joint. Compare the difference in resistance induced by first and second technique.

Ill have to try that when im back home. That qill take about 30 mins.

[Conrad Hoffman]

That will work if you have a 6 1/2 digit meter and the resistance value is in a "good" location on one of the scales. Better is to make up a 4-resistor equal value bridge and look at the null.

[Spikee]

i will do the resistor matching using a wheatstone bridge but that is for a later date...
I just checked my Agilent U8001A psu (in calibration) versus my rigol DM3058 and my newly acquired Fluke 87 III .

The maximum set resolution of the psu is 2 after the decimal point but that will be good enough indication till my dmm check arrives.

Agilent U8001A set voltage: 01.99V
Fluke 87 III:    1.9883V (lsb from 2-> 4)   
Rigol Dm3058: 1.98894V (average 2 minutes) (have to hook up to the pc to see the full 6 1/2 digits)
The test leads are both equal length so i can assume the Fluke is the one that has drifted a little.

I am testing the non soldered resistor vs soldered resistor and soldered with heatsink and without now.
       

[ftransform]

Are you gonna go after some kind of oil immersion? I know most people here say it is a bad idea but I think it would be cool because no one has ever done it at home that I know of.

AS ghetto but good quality moisture sealing why not put your tupperware (with oil and resistors and whatnot) into a big mylar bag with a desiccant packet? Even heat seal it maybe.

[Jay_Diddy_B]

Hi Group,

Remember if you are needing the ultimate in precision, you can not use any old cable to hook up the instruments, you should use low thermal EMF cables like these:

http://www.pomonaelectronics.com/pdf/d5291A_102.pdf

If you have a sensitive meter, you can see them drift if you warm the input terminals with your hand.

Check these application notes from Linear Technology:

http://cds.linear.com/docs/Application%20Note/an09f.pdf


This other application note talks about how to get high accuracy using DACs and ADCs. By using a DAC you remove the need to match all the resistors, the manufacturer has already done that for you.

http://cds.linear.com/docs/Application%20Note/an86f.pdf



Jay_Diddy_B

[Spikee]

Measurement results:
https://dl.dropbox.com/u/100819959/resistor%20solder%20drift%20test.pdf

Interesting findings, If you use pliers as a heatsink the resistor wont show a drift (at 5 1/2 digits measurement)
I could try hooking it up to the pc and read the 6 1/2 digits but i don't find that necessary.

@ftransform
I canceled the plans to do the mineral oil method because it would not deliver the results i want to achieve. It would try it if i had any mineral oil laying around ...

@Jay_Diddy_B
My test leads are exactly like those ones except they are not gold plated. Ratings : Cat III 1000V / 16A
The application note 86 has some good inf about kelvin varley dividers
And my favorite programming language is solder.

It is time to sleep now =/

[ftransform]

any progress?

[Spikee]

Still waiting for the switches . . . They probably are at customs :€

[cellularmitosis]

Any progress on this?  I've been very interested in building a kelvin varley divider myself recently.  Has anyone else on the forum built one?

By the way, Conrad, thanks so much for your home metrology lab articles!  Inspiration!