HW Sullivan Volt Ratio Box from 1964

[trobbins]

I'm helping a friend check his VRB.  Luckily he has the original standards lab folders on the unit that cover the first 2 decades of its life, and make for an interesting read.

It started life in 1964 as a HW Sullivan Type T 2100, with 200 ohm/volt sections and ratios of 2, 5, 10, 20, 50, 100, 200, 500 and 1000V to 1V and a 0.001% accuracy at rated voltage and 20degC.  Original British NPL calibration report indicated low ppm error and uncertainty, up to 7ppm error for 500V and a 5ppm uncertainty for 200 to 1kV ranges.  It had repeated cross comparisons over the next 2 decades with two other volt boxes of the same type in a defence standards lab, and showed at worst possible low ppm drifts.  Testing also showed at most 2ppm change from 1/10 to rated voltage (ie. 5W at 1kV), and 200,006.3 total.  Local test equipment used was typically a Guildline DCC and a 100 and 10k references cal'd to 0.1ppm, but not much detail of that in the notes.

There is no provenance after 1985, but it then likely spent a few decades in a test section at an electronics company, and then literally a farm shed.

I've only quickly confirmed that the section resistances are nominal with a handheld dmm, but first I need to clean the external surfaces, and check the oil for moisture and IR.  When it landed in Oz it was filled with high voltage transformer oil by the local electricity supply lab, and as far as the lab notes go the oil is likely original.  The oil shows only a slight colouration if any and is to a nominal fill level by the dipstick.  Apart from a crackle test, what in-situ IR testing has anyone done, and is that appropriate?  The original oil spec was 2x 10^12 ohm, and a test after about a decade measured it at about 1.1x 10^12 at 500V.  I can IR test from the 0V tap to enclosure screen although my handheld only goes to a 2x 10^9 reading, so if it is more than that then I'll have to break out the vintage BPL 175LZ Mk2 megohmmeter which goes to 5x 10^12 at 500V.  I'm not keen on replacing the oil at all, and certainly not initially, and would prefer to leave that decision to later.

The two thermometers each have 0.2degC minor gradations, with a 15 to 25C span.  I'll let them settle, but they both were showing the same value so that is a good start too.

The step values of 200, 200, 600, 1k, 2k, 6k, 10k, 20k, 60k, 100kohm should make it relatively easy to bridge compare groups of steps against some balanced resistor pairs I have using a 6-digit meter for null measurement and using polarity swapping, so that is likely how I will start.  If all goes well then it may not end up being used as a volt ratio box per se, but more for absolute resistance cross-comparisons and also for 1, 10, 100 and 1000 ratio checking of Wheatstone bridges.

Tim

[trobbins]

Update is that it looks like all sections are within 0.05% on my uncal Keithley 197, and sections from 6kohm up are better than 0.01% based on comparison to various Fluke fixed resistors.  So I'll move on to a better form of measurement, although that means moving it to another location, which is a pain as it weighs at least 25kg.

Not unexpectedly the IR between metal enclosure and elements has degraded, and now is circa 1250Meg at 500V, so that may show up in the next round of measurements.  I'm checking what oil is available locally for retrofil given it holds circa 20-25 litres.  At least I can take some internal photos if I change the oil, and assess if there is any sludge or corrosion.

I'm still not sure if a simple IR test from enclosure to elements was the method used, given that transformer oil testing appears to have typically been a voltage withstand test.

[Gyro]

...  it holds circa 20-25 litres.

 

EDIT: That's an awful lot of bottles of Laxative (potentially)!   https://www.eevblog.com/forum/metrology/oil-for-a-leaking-tinsley-standard-resistor/

[trobbins]

Yes the metal enclosure portion is 410Lx380Wx195Dmm so 30 litre - a veritable swimming pool.  But an expensive pool to fill by the first price I got for a 20L can.

Thanks for the link.  I'm just starting to rationalise the options for the volt box.  Draining the oil seems a reasonable option, as that would remove the issue of leakage, and the unknown condition of the internals.  The box is rated for 5W dissipation at 1kV, which wouldn't be an issue if just used for resistance or low voltage ratio measurement.  And given the comments about oil replacement periods down at 2-5 years, any requirement for oil is an onerous burden going forward.

[alm]

The oil won't be just to improve power dissipation, but also to keep all resistors at (almost) the same temperature to reduce the effect of their temperature coefficients on the voltage ratios. So I expect that even at low voltages draining the oil will degrade performance as a voltage divider.

[trobbins]

Totally agree - the option to fill with oil is still there.  Still a long way to go in the assessment phase, which will start with old oil as a benchmark.

[Gyro]

I'm sure that the industry has methods of regenerating transformer oil, maybe with dessicators (add some silica gell?), treating under vacuum etc. It might be an alternative to replacement as long as the oil doesn't cause corrosion.


Edit: https://globecore.com/oil-regeneration/transformer-oil-regeneration-theory-technologies-and-equipment/ (there seem to be quite a lot of search hits). Fullers earth looks to be a possibility (cat litter?).

[trobbins]

Chris, there is a local transformer oil processing facility, but it is industrial scale for use by the electricity distribution industry.

I rechecked the insulation resistance between chassis and resistor elements and measured 8,000 Megohm at 500Vdc using a BPL RM175 LZ Mk2 megohmmeter and reference resistors to confirm.  That indicates leakage current between resistor segments is likely to have a negligible affect on my testing down below 20Vdc, given my poor-mans test setup.

I started with a 4W resistance measurement using an uncal Keithley 197, which gave measured values of 199.987, 199.965, 599.88, 999.79, 1,999.55, 6,000.4, 10,000.6, 20,001.6, 60,000, and 100,000.  The 20k VRB segment is effectively the same as two Fluke 19.999k 0.05% resistors (from a 895A Differential Voltmeter - Kelvin-Varley board assembly - that I had match selected and placed in a 4W reference box) which measure 20,001.5.  So it's likely that all sections are still <0.05%, and some may be <0.01%.

I then used a CCS and a DVM to make measurements across adjacent resistor segments.  The setup used 10uA, 100uA or 1mA CCS from an uncal HP3497A, connecting as 4W and with a nominal working span up to 20V, but I didn't exceed about 15V and used two CCS values for comparison where possible.  Voltage measurement used a Picotest M3510A 6.5 digit uncal DMM on either the 1V or 10V range, with leads moved from across one resistor segment to the next (within the span of the CCS circuit).

For starters I compared the two Fluke 19.999k 0.05% resistors with the 20k resistor segment of the Sullivan VRB, and there was no measurable difference (ie. <0.001% resolution difference) with the 0.1mA CCS and 10V range.  So that was consistent with the Keithley 197 measurement.

I then compared two adjacent VRB resistor segments using the same CCS setting and voltmeter range, to minimise measurement condition variation.  The measured resistance ratios were compared to the stated VRB voltage ratios between adjacent segments (ie. 1, 10/6, 2, or 3).  The % error in measured ratios were:
segment ohm      error
200 to 200    0.0110%  0.0125%
200 to 600    0.0028%  0.0032%
600 to 1k            0.0037%  0.0027%
1k to 2k            0.0010%   0.0095%
2k to 6k            0.0392%   0.0327%
6k to 10k            0.0380%   0.0393%
10k to 20k            0.091%
20k to 60k            0.391%
60k to 100k    0.389%   0.395%

Two error values relate to two setup conditions, such as 1mA 10V and 0.1mA 1V, or 1mA 1V but through more than two segments.  This is not the measurement technique normally used for a VRB, but was simple to set up, and appears to show up the influence of the 10Meg input resistance of the Picotest DVM.

I may get keen to set up a bridge next, and compare 200:200, and 200+200+600:1k, and up to 100K:100k, using the two Fluke 19k999 resistors as 1:1 ratio arm references, and see if any added arm resistance needed for balance is consistent with the resistance values measured by the Keithley.

All good fun!

[Gyro]

8G at 500V sounds pretty good, no I don't think it would have any effect down at 20V with those resistance values. I suppose the only possible harm old oil can do is any acidic corrosion - I don't know if litmus paper would work with oil. I suspect acidic breakdown only occurs if the oil is run hot and with high (as in transmission) voltages, but I'm no expert.

Indeed, it sounds like fun!

[trobbins]

Yes, it is uncertain how much shunt resistance across individual VRB resistors is occurring due to the oil bath itself, or from creepage along the top insulation board between terminals, and how a bulk measurement of IR to the metal enclosure relates, as the IR measurement may be dominated by one or other paths.

I just did a simple set of bridge measurements as the next step in assessment.  I used my two Fluke 19k999 matched resistors as 1:1 ratio arms in a bridge to compare 1:1 ratios from the Volt Ratio box.  I may get a chance soon to measure the 19k999 resistor absolute values with better confidence, but for now I am assuming they are the same.

The volt box 1:1 ratio arms used in the bridge were 100k:100k, 20k:20k, 10k:10k, and 2k:2k, where one of those VRB arms is a single resistor, and the other arm is the series of lower valued resistors down to the COM terminal.  The bridge was energised from a 24Vdc battery through a polarity reversing switch, and the galvinometer was a convenient handheld Aneng 8009 which is battery operated and has resolution to 5uV.  The links between the two halves of the bridge were measured at 90 milliohm and within 0.3 milliohm of each other.  Balance was met by padding one or other arm of the VRB with a low value fixed resistor, and achieving <=5uV reading for both polarities.  VRB oil temperature was 16.8C, and Fluke resistor pair temp was similar, so tempco of all sections of bridge come into play, but for now are assumed zero (the Fluke pair had +0.5ppm/C and -0.5ppm/C batch markings, and the VRB section tempco's may well be more so some tempco influence is likely).

The 100k:100k VRB arms (COM-500V-1000V terminals) balanced with 660 mohm in the 1000V arm.

The 20k:20k VRB arms (COM-100V-200V terminals) balanced with 220 mohm in the COM arm.

The 10k:10k VRB arms (COM-50V-100V terminals) balanced with 22 mohm in the COM arm.

The 2k:2k VRB arms (COM-10V-20V terminals) balanced with 22 mohm in the 20V arm.

I didn't try to check the lower VRB 1:1 ratios (COM-1V-2V, and COM-5V-10V) - perhaps when I get a matched pair of reference resistors of value around 200 to 1k ohm, and can setup a 20C environment.

So, it looks like the VRB resistances still have a tight ratio of circa 0.001% from the above results.  I also enquired further on the VRB history, and it was moved from the original Standards Lab in Adelaide to a Melbourne Lab in 1993, and has been dormant since then, except for the last 6 months.  So it has led an easy life.

Vintage test results did not cover individual resistance tempco, so I may try and determine the approximate tempco of the VRB sections next using just Keithley 197 resistance measurement and a heater pad in a tub that the VRB is in, but given the thermal mass, this may take some time.