Advantest R6581 8.5 digit DMM mini teardown/repair

[TiN]

Mm, a cal report to cruch on. First question I would ask, what is accuracy column exactly mean? Voltage 10V point at 0.2ppm? 10kOhm at 0.5ppm? Those are not even 24 hour specs, so  ?

R6581T I had was very dodgy on ohms at ranges 100kOhm and higher too, and running ACAL always gave different results. Then I have given away the meter to a less fortunate volt-nut, thus resolving the issue and marking R6581T as not worthy to have power of 8.5 digits,

[dietert1]

TiN, we can read what you did to that R6581T in the thread about your temperature chamber. I have been running the R6581T in a temperature chamber, too. I think the criterion is the RMS value of its internal temperature sensor. Inside the chamber it comes down to 0.01 °C (readings every 5 seconds). This is about the power supply of the R6581A floating section, in the box near its fan.
After some days at 23 °C it settled to be an 8 digit meter without ACALs. Residual noise came down to about 200 or 300 nV rms including noise of the 10 V reference (DUT). The linearity issue is one of the bugs to be fixed (read above). So it depends a bit on the user what it does.

Regards, Dieter

[leighcorrigall]

Mm, a cal report to cruch on. First question I would ask, what is accuracy column exactly mean? Voltage 10V point at 0.2ppm? 10kOhm at 0.5ppm? Those are not even 24 hour specs, so  ?

R6581T I had was very dodgy on ohms at ranges 100kOhm and higher too, and running ACAL always gave different results. Then I have given away the meter to a less fortunate volt-nut, thus resolving the issue and marking R6581T as not worthy to have power of 8.5 digits,

Yeah, it seems like a sketchy instrument...

Do you have an English version of the specifications/data sheet? I only have a Japanese copy.

[bdunham7]

Looking at the 'as found' data, if the meter was in good ohms calibration at some prior point, the discrepancy now on all four ranges could be accounted for by a constant ~500pA worth of leakage or excess bias current.  Since there appears to be 10V test voltage on all of the affected ranges, the cause could be anything in the circuit that would have a consistent leakage current at 10V.  I thought about looking at the low current ranges to see evidence that a bias current might be involved, which does not appear to be the case, but I did notice that they listed the test results for the 100nA range in uA--I hope that's a typo, but it seems odd that their systems would make such a mistake.

[ramon]

Oh, how sad. Look those poor R6581T owners! They probably should burn their devices into ashes, or at the very least hammer them to pieces with fury. Definitely they must all be destroyed! And the 'R6581T' name should never be spoken again into the temple of 8.5 digits DMM.

I'm going to tell you an anecdote.

It was around February/March 2018 when we (TiN and me) arranged a meeting at my home. TiN received back his 10K resistor (SL935) from USA calibration recently (maybe Dec 2017?). The R6581T was running for some hours before the meeting. He meticulously cleaned the banana jacks with Deoxit. (Yes, he brought a complete calibration kit to my home. Thank you Illya!). Because he didn't bring with him his own 4-wire cable, he carefully tested my 4 wire cable (after some testing, he considered that it was ok). He connected the SL935 10K to the R6581T, and a USB-GPIB cable to his 17" laptop. As you all know, he likes to have his own scripts with everything automated.

He run the 4W resistor test script. At first, the 10K measurement was way off. Something was wrong! He found that the value was increasing ... It was self-heating ! We indeed were using the 4W high-power mode and the resistor was getting hot.

We changed settings to 4W Resistor LOW power mode, and waited some minutes. Now, this time the SL935 10K measurement agreed to 1ppm. It was spot on.

His SL935 was just recently calibrated (less than 3 months), my R6581T last external calibration was on 2004 (14 years).

[bdunham7]

He run the 4W resistor test script. At first, the 10K measurement was way off. Something was wrong! He found that the value was increasing ... It was self-heating ! We indeed were using the 4W high-power mode and the resistor was getting hot.

We changed settings to 4W Resistor LOW power mode, and waited some minutes. Now, this time the SL935 10K measurement agreed to 1ppm. It was spot on.

What are the applied currents/voltages in the high and low power modes? 

[leighcorrigall]

He run the 4W resistor test script. At first, the 10K measurement was way off. Something was wrong! He found that the value was increasing ... It was self-heating ! We indeed were using the 4W high-power mode and the resistor was getting hot.

We changed settings to 4W Resistor LOW power mode, and waited some minutes. Now, this time the SL935 10K measurement agreed to 1ppm. It was spot on.

What are the applied currents/voltages in the high and low power modes?

Hi bdunham7,

You can read the attached Section 3.7 that I mostly completed last night at 1:30 AM on my vacation 

I took a screenshot of the table on my MacBook too.

Regards.

[bdunham7]

OK, so essentially 10V on HI and 1V on LO, except the 1G range is always 10V, with the currents calculated accordingly.  So according to Fluke 742A specs, HI would be causing a very minor, acceptable error on the 10K range and then essentially no error on the ranges above that.  I would presume calibration would be done in the HI range.

TiN's rather unusual resistance reference referred to above is a different animal than the 742A series.

So I think you may have a half a nanoampere going somewhere--good luck finding it!  It might be instructive, if the cal lab is cooperative, to remeasure the highest 4 ranges in the LO mode.  If it is leakage in a set ratio to the voltage, or highly dependent on the voltage, you would see the 1G range unchanged, the next two ranges would shift towards (or even beyond) the correct value and the 1M range might even read high.

[dietert1]

One sanity check is a DC voltage measurement with only a 1 GOhm and a little HV cap over the inputs. For our R6581A the result was well below 1 pA.
If Mickle T. is right in that no user setting is used during external calibration, then maybe the instrument is trying to do it in 10 PLC mode and one should use a HV cap parallel to improve results. Still it isn't guaranteed the capacitor will be charged properly at the right time. Or use a guard box.

Regards, Dieter

[leighcorrigall]

Thank you.

If and when Texas HQ responds back to my rather lengthy email, I will include these details. They might get impatient with my requests, but this is what I am paying them for.

I believe I have a few precision resistors and capacitors that I can reliably use to diagnose the high-resistance ranges of the DMM when I get it back. 

Hope is not all lost. At least it works in most conditions and I can use it as a 'transfer standard' for the usual ranges.

[bdunham7]

Hope is not all lost. At least it works in most conditions and I can use it as a 'transfer standard' for the usual ranges.

Not to be dismal, but IMO you do need to understand where the problem is before you can really trust it in any range, including DCV and DCA.  Calibration constants can hide minor flaws but you don't necessarily know how that might affect intermediate values as opposed to calibration points, nor do you know if the issue is stable.  My guess is that the other ranges are OK, but if I were you I wouldn't relax until you at definitively isolate the issue to something in the resistance section.

This sort of struggle is hard enough at 5.5-6.5 digits and 500nA test currents.

[Kleinstein]

One weak point of the R6581 design is the switching at the input (e.g. in AZ mode) under some conditions there can be quite some charge injection / current pulse at the input. Especially for a short integration this may act like additional bias current / an input resistance lower than the normally expected value. It may very well still look OK at zero with the simple test with a high values resistor, but the input current at some 5 V or 10 V may be quite different.
Extra input leakage can happen at higher voltage (the switching JFETs see a higher voltage with a positive input).

A possible way to check the input current is to record the drift if the read votlage with a small (e.g. 1 nF) low DA capacitor at the input. Charge the cap to suitable starting points (e.g. +-10 V, maybe +-5 V and 0 V and than record the drift. The integration time (e.g. 1 PLC vs 10 PLC) and AZ setting can make a difference from charge injection. The input bias is drift rate times capacitance (including DMM internal capacitance). The input current is not necessay constant or a simple linear function of the input voltage.

This is a simple check, I recommend before sending a DMM with high Z input (e.g. > 1 G) to calibration or when getting a new / used one.  Even without computer read out one can get a pretty good idea of the dirft rate and thus the input bias.

[leighcorrigall]

...

First question I would ask, what is accuracy column exactly mean? Voltage 10V point at 0.2ppm? 10kOhm at 0.5ppm? Those are not even 24 hour specs, so  ?

...

Illya is correct. Where is the accuracy information coming from? The resistance specifications do not match the calibration accuracy spreadsheet... 

I haven't fully evaluated everything, but it appears that the 10 MΩ and 100 MΩ may actually be in range. 

[leighcorrigall]

Out of curiosity, what do you do with your temperature measurements?

These temperature measurements simply accompany long-term drift measurements 

Attached is an English translation of Section 15 - Specifications of the original ADCMT R6581(T) manual from 2007 (pages 621 to 635).

Dumb question / sanity check:
Is the "23 ± 1 °C" for the 24-hour specifications referring to the ambient room temperature or the internal temperature of the instrument?

[branadic]

Ambient temperature. Internal temperature is somewhat higher

-branadic-

[dietert1]

The difference is about 14 to 15 °C for our machine.
And the internal temperature sensor is slow in comparison to TC of the DC Volt measurement, so it isn't easy to use for TC compensation. Best use the R6581 at constant temperature.
For our machine TC is about -0.1 ppm/K, so in order to get 8 digits one needs +/- 0.1 K, and that actually works. Noise is less than 0.02 ppm RMS (200 nV) with 100 PLC. This is a conservative limit, including some noise of 10V test reference.

Regards, Dieter

[leighcorrigall]

Thank you, Branadic and dietert1.

I spoke with the TRANSCAT representatives today. Looks like they recognize that they made critical errors with their tolerances and some units. They have at least acknowledged them for now...

The temperature variation for the calibration testing was recorded at 23.47 °C (as found) and 23.13 °C (as left). Temperature variance could be a significant factor in the 1 GΩ resistance measurement, which is the only range that is not within tolerance (1.6 % error) according to my own calculations based on the 2007 ADMCT manual. The magnitude of the error is highly suspicious. When I get the instrument back, I will record all of the previous settings for each mode (DCI, DCV, OHM) to determine what configurations (PLC, HI/LO Power, AZERO, INT CAL, COMP, etc.) they used. Hopefully, the settings will reflect the measurement conditions. From there, I can assess whether or not the instrument was used properly to determine the tolerances.

The staff in Texas HQ are swamped with work due to a COVID outbreak that has resulted in the loss of one of their members recently. 

It is going to be a while until they can validate their work and return my instrument. At this point, I have a fair bit of confidence that my 'toy' is working properly. I have an Ohmite RX-1M Series resistor at 1 % tolerance (measured as 1.0025E+09 Ω by R6581T, 1.0025E+09 Ω by R6245, and 1.0038E+09 Ω by R8340 on 2021-03-07) that I will use to check the R6581T 1 GΩ range. Luckily, I tested all the measurement ranges before I sent the DMM to TRANSCAT. Since the R6245 and R8340 have not been changed, I should be able to compare them with my R6581T using the Ohmite resistor. If the R6581T measures comparatively well with the other instruments, like it did before, I will know that TRANSCAT has measured the 1 GΩ range incorrectly.

[bdunham7]

I spoke with the TRANSCAT representatives today. Looks like they recognize that they made critical errors with their tolerances and some units. They have at least acknowledged them for now...

Forgive my obtusity, but what error have they made?  In the three relevant ranges, the only thing that raised an eyebrow for me was the TUR on the 10M range.

[leighcorrigall]

I spoke with the TRANSCAT representatives today. Looks like they recognize that they made critical errors with their tolerances and some units. They have at least acknowledged them for now...

Forgive my obtusity, but what error have they made?  In the three relevant ranges, the only thing that raised an eyebrow for me was the TUR on the 10M range.

If I did the math correctly (assuming LOW Power, TCR, and EXT CAL):

9.99441200 MΩ < 9.99856300 MΩ < 10.0056840 MΩ (PASS)
99.4964093E MΩ < 99.8658000 MΩ < 100.517131 MΩ (PASS)
994.575354 MΩ < 984.106400 MΩ < 1005.60265 MΩ (FAIL? possibly human error)

This is not the same as what TRANSCAT provided (assuming HI Power only):

9.99948800 MΩ < 9.99856300 MΩ < 10.0006080 MΩ (FAIL)
99.9555700 MΩ < 99.8658000 MΩ < 100.057970 MΩ (FAIL)
995.076600 < 984.106400 MΩ < 1005.10140 MΩ (FAIL)

TRANSCAT will not reveal the following:
- power (HI/LO) level, which dictates the tolerances
- integration time

Additional information:
- The instrument successfully measured an Ohmite RX-1M 1 GΩ with 1 % tolerance (1.0025 GΩ) prior to replacing R200 and R234.
- I left the instrument running for months before sending the instrument for calibration. It was tested with the resistors I had available. It was a quick test to determine if there was anything unusual before sending it in.
- The manual is not available in English and the calibration lab did not have access to translations to configure the settings.
- My guess is they were using whatever parameters the instrument was set to and quickly measured the ranges. I will find out what these settings are once the instrument is sent back.

It is difficult to believe that the instrument is dysfunctional based on the information I have presently. Fingers crossed though. My only concern is that the resistor replacement mentioned above could have somehow caused the 10 MΩ to 1000 MΩ ranges to fail. If so, I suppose I know where to begin diagnostics. My question is how could those resistors impact the high-resistance range only? 

[Kleinstein]

I don't see how a change of R200 and R234 could effect the high resistance ranges only.  The effect would normally be small, like a minute change in the INL and drift.  So it may effect the ACAL part for 1 V. The high resistance range don't really care about low ppm's - so even a rather poor R200 / R234 would not be visible for the higher resistance ranges.

The high resistance ranges are more sensitive to the ciritcal OP in the current source (U404), leakage of D402  (at the protection), Q402-404. The general input bias current could be an issue too.  As the gain is set via ACAL, a problem in the current measurement portion (e.g. leakage around U503) could also effect the high Ohms accuracy.   It does not have to be a broken part, but deposits on the PCB could also cause leakage. Some leakage only causes offsets that are corrected anyway.

It is odd, that they did not ask for a manual if they don't have one.
I would consider that using 100 PLC would be the obvious choice, even if you don't have a manual. Anyway it would not make a difference if they use 10 PLC and than averging, as 100 PLC is also just doing that internally.
The Ohms power setting should not effect the higherst ranges - these are allway the same low power, limited by the availabe voltage.

[bdunham7]

If I did the math correctly (assuming LOW Power, TCR, and EXT CAL):

This is not the same as what TRANSCAT provided (assuming HI Power only):

TRANSCAT will not reveal the following:
- power (HI/LO) level, which dictates the tolerances
- integration time

I suppose having the manual only in Japanese is quite an issue.  Are there any Japanese forum members that might help translate a page or two?

I would have thought that the calibration procedure would use HI power, but of course that doesn't mean they are using the HI/100PLC settings to measure the 'as left' reading.  However, a thought did occur to me--is it possible that the calibration procedure does both HI and LO and there are two separate sets of calibration constants for each power level?

Also, although I have no experience with that DMM, or 8.5-digit DMMs in general, but typically if a machine reaches its limits for calibration constants, it gives you an error.  So either the DMM should enter a constant large enough to give you a correct reading--no matter how far off it is to begin with--or should not enter any constant but instead tell you that the baseline reading is out of limits.  To enter a value and have the machine accept it and then not measure accurately later would be indicate that something is going wrong.  I have had meters that do accept a calibration point, but then have either instability or gross errors at points other than the calibration point--which is why I harp about not calibrating broken meters.  But to have it accept the entry and then immediately be off at the same point seems weird.  If the readings are stable, then the only reason I can think of for this to happen are that the readings after the calibration point entry are being taken with different parameters than during calibration, whether that be power level, PLC or something else.  Or, possibly the meter doesn't work the way I think it should--and I can't read Japanese.

[leighcorrigall]

I just got off the phone with the calibration technician. He said that he tried everything to make the R6581T measure within tolerance:

- 2-wire mode instead of 4-wire for the 10 MΩ, 100 MΩ, and 1 GΩ range (they will update the accreditation data to reflect this)
- checked both HI and LO power settings (not much difference)
- used 100 PLC for the whole procedure
- OMH-COMP OFF

Apparently, the technician had no problem with the R6581T because it is similar to the Fluke 3458 and 8508.

He mentioned that while testing, he noticed something very strange about the -1 V reading on the 10 V range. He had to calibrate the DMM three to four times to get the instrument to read correctly. This might be key to diagnosing the problem... EDIT: This sounds like ACAL R230, which is located right next to R234.

I think I must have done something wrong on my end. Maybe I overheated a component while de/soldering or left part of the board contaminated enough to have poor readings at the high-resistance range. 

EDIT: Change history:
1) Branadic binding post adapter with leads soldered to posts
2) R200 - easy replacement with solder sucker because the feet were straight
3) R234 - particularly difficult to remove with a soldering iron, solder sucker, and eventually wire cutters
4) K006 - followed the picture and heat shank a cover over the green wire
5) C215 - easy to overheat, but should not be a critical component


I am going to get an estimate for the cost of repair if possible. Might as well have the instrument come out completely calibrated.

[Kleinstein]

AFAIK the 6581 uses ACAL similar the the 3458. So there are no real cal constants to enter for each Ohms range, let alone seprate values for the high / low power modes. Ideally there may be some extra tests and constants  for the very high ohms ranges modes to account for leakage, but I somewhat doubt they have implemented this.

The CAL steps at these modes are tests, if the performance is acutally within the bounds. AFAIK there is no individual adjustment, just 10 V and 10 K.

In most cases a calibration is a check only and there is no adjustment, maybe except of zero offsets (e.g. 4 wire ohms zero and voltage zero) that is also user adjustable independent of calibration anyway.

Trouble at -1 V may be especially bad: the ACAL procedure uses -1 V to get the range for the 1 V range and lower. If the porblem is with -1 V at the ADC it would also effect the 1 V to 100 mV range step. So the readings at -10 V, -1 V and 0 V are the 3 most imortant ones.

[bdunham7]

The CAL steps at these modes are tests, if the performance is acutally within the bounds. AFAIK there is no individual adjustment, just 10 V and 10 K.

OK, that explains how you end up with a wrong reading but no error message.  However, I can't square that exactly with his as-found vs as-left data.  The 10K range was adjusted down slightly, apparently, but the 4 top ranges went up.  The 1M range actually went up enough to get back into spec.

[Kleinstein]

There are 2 steps in the calibration process with adjustment. The first is to adjust the constants for the 10 V and 10 K resistance range. This is normally  only done rarely, e.g. for a new calibration after a major repair / change, or if starting well out of cal.

The 2nd step is the ACAL procedure, the internal measurement of the internal gains and ratios of the shunts and current sources. This process is normally repeated relatively frequent, like once a week or one a day, e.g. before a critical measurement sequence. Normally the ACAL part should be rather repeatable and at the same temperature give very similar results. Over time is may slowly change and this is why it is repeated.
The high resistance ranges are about the most sensitive parts in the ACAL process as it used very small currents and thus react to internal leakage currents. Leakage currents can change with temperature and humidity - the humidity effect may take quite some time.
Repeated ACAL under stable conditions would be a point to normally test - the change should be rather small, like less than 1/2 the 24 hours specs.
This may not be part of the normal CAL procedures - though in my view it should and there should be test limits for this. On could consider this an extra self test, as it does not need any extra equipment.