Advantest R6581 8.5 digit DMM mini teardown/repair

[Kleinstein]

The internal calibration checks the gain of the 1 V range at -1 V. So ideally the -1 V point should be good by design. The +1 V is no accurate by design - it is more a test for the linearity for the 1 V and 10 V range. Even though the possible error is visible when reading 1 V, the INL error is kind of the average from the positive an negative side.

A full linearity test would need a really good instrument (e.g. high end calibrator and / or good 3458).

The turn over test only checks a part of the INL, but is still relatively easy. Ideally one needs good switches for the polarity reversal without much extra thermal EMF error. 

[dietert1]

Yesterday i replaced U107 OP07 by an OPA140. I think this will help for accuracy, especially in DCV, 1 V range.
The voltage divider R100 has an output impedance of about 4.4 KOhms in that range. This adds to the switch resistance of about 1 KOhm of U101 MAX327 (present with all three gains). As the OP07 is specified with input currents in the nA range, there are unknown offsets of about 1 uV to 5 uV. This order of magnitude estimate gives 1 uV = 10 ppm in the 100 mV range, 5 uV = 5 ppm in the 1 V range and 1 uV = 0.1 ppm in the 10 V range. These contributions are neither known very well nor stable.
With the pA input current of the OPA140 these errors are essentially gone. For stability i added a 100R resistor between the OPA140 output and the guard network. I measured 1180 pF between guard and Gnd (R6581T power off). A low impedance at instrument input may increase the capacitive load when powered on. I checked this configuration to be stable with all three gains of the fast channel.
Another small step of making the R6581T a 8.5 meter.

Next i will try to add another MAX327 between U101 and the amplifier channel inputs to run the unused amplifier channel with gain=1. Let's remove random signals in this section of the instrument and clipping with ground current injection at D120.

Regards, Dieter

[Mickle T.]

Replacing U107 and U504 - is the one of the first things I did with my DMMs. But in my case there was no any measurable improvement in low-V ranges. Mainly because OP07D is a low-noise / low-Ib version of OP07 and already has an excellent input current compensation (or may be due-to pre-selection in production).

[dietert1]

The R6581T has multiple design problems and removing/modding any of them won't show a significant improvement until you solved the last one. As long as one or two of them are left, it remains a 6.5 digit meter. Selecting an OP07 for pA input current would be more like a piece of art and might depend on illumination. The OPA140 is better than needed "by spec" and it has a similar power consumption. I didn't use a TL071 as it has about 20 times higher low frequency noise amplitude.

For U504, maybe one could add offset voltage trim. In the instrument i am working on, its offset voltage is about 70 uV. Or replace it by an OPA189 chopper. That mod probably destroys external calibration of lower voltage ranges.

Regards, Dieter

[Kleinstein]

The input bias from the OP07 U107 would add some offset voltage. However this would not be a problem as the offset would be essentially constant and is corrected in the normal auto zero cycle. The point would be more the common mode input resistance, and here the OP07 is quite good (some 160 GOhms typ - may vary with manufacturer) and here the nonlinear part of the resistance could have a small effect. The linear part only changes the internal gain a little and this gets corrected with ACAL.

AFAIR the incentive to change U107 was to make it faster and this way reduce the gate current spike a little. There may however be stability issues.

Another odd point is to use the output of U107 for the guard also for circuit parts at the very front. For some parts I would consider the output Of U009 more suitable.

U504 adds  a little to the noise and offset / drift before AZ. Here the offset and drift of the OP07 is usually considerably better than the main amplifier. The noise is also not that bad, though there are lower noise (the relevant frequency would be something like 2.5 Hz) ones available.
A change of U504 should normally not effect the calibration (gain adjustment for the 10 V range). It would however effect the internal calibration and possible linearity corrections. Besides noise, the ouput cross over distortion of the OP07 could habe some effect on the linearity in the 100 mV range.  A formal calibration with the paperwork would normally be considered void after such a change of cause.
One could check the possible nonlinear effect by loading the output of U504 when reading 0 in the 100 mV range.

[dietert1]

After removing the original JFET MUX with the strange feedback loop where switching depends on the switch output the guard driver speed does not matter anymore. For me the point was to remove the offsets caused by input offset currents. Those currents are extremely temperature dependent and they are so easy to remove that one could call the placement of OP07 a design error. There may be historical reasons, though.

Here i have a scope image of fast mode with 1 msec integration time and autozero, showing integrator output and guard driver output. Input voltage is 3.5 V.

Regards, Dieter

[Mickle T.]

Please! Can anyone provide the diagnostic data of R6581 True-RMS board?

[bobAk]

Amp_off +0.0008 - 0.0005
Vca_off -0.00226
Ac_ref1 -9.81724
Ac_ref2 +9.81230
Abs_dead1 +0.17970
Abs_dead2 +0.17504
Abs_off1 -0.17932
Abs_off2 +0.17510
10vf_1 +1.86881
10vf_2 +1.91763
10vf_3 +1.91175
10vf_4 +1.86854
Some readings are drifting.  Everything related to 10vf_X, ref is stable to the last sign.

Add. You can measure noise at the 100 DC volt range with the input shorted? Config 100 nplc 8.5 digits az on.

[Mickle T.]

Thanks! I am trying to recover a completely dead R6581 with multiple resistor failures. At the moment, 7 open-circuit resistors have been found. The device comes to life and successfully passes self-test and internal calibration. However, I cannot find the reason for the huge temperature zero offset of the AC measurements.

[dietert1]

I think this multiple failure is a bit unusual, except somebody threw the instrument or it got hit hard. What does the metal enclosure look like?

Regards, Dieter

[serg-el]

Cumulative Measurements R6581

[Mickle T.]

I don't think these failures are due to mechanical stress.

[dietert1]

Mechanical stress may also happen when you freeze the instrument in an airplane cargo bay. I mean without traces on the metal enclosure. Did you think about the reason of the failures? Somebody connected the instrument to 220 VAC while setup for 110 VAC?

Regards, Dieter

[Kleinstein]

Exteme cold temperature could be an issue, also without a plane. Some chemical problem is also possible, maybe overly aggressive cleaning.

It is still strange to see resistors scatttered all over the circuit to go bad. This really does not look like a damage from overvoltage.

[bobAk]

Cumulative Measurements R6581

do you have raw data?

[dietert1]

All those resistors are driving zeners. If the input is unregulated supply voltage, it does look like damage from overvoltage. Anyway overvoltage is a possible explanation for multiple failure.

Regards, Dieter

[serg-el]

Cumulative Measurements R6581

do you have raw data?

You are welcome.  Raw measurement data of short input R6581 with AZ on and AZ off with different PLCs.
https://cloud.mail.ru/public/miPE/s2kbDdSre

[kleiner Rainer]

All those resistors are driving zeners. If the input is unregulated supply voltage, it does look like damage from overvoltage. Anyway overvoltage is a possible explanation for multiple failure.

Regards, Dieter

I had problems with SMD Resistors not being able to stand inrush currents. The cause was the way those resistors are trimmed: the resistance paste is screen printed onto the substrate and then fired. The value of the resistance paste is selected so that the value is always lower than the nominal value of the resistor. Then the resistance is trimmed close to nominal value by cutting into the resistive layer between the two contact areas with a laser or by sand blasting. Depending on the tolerance of the fired paste, there will remain only a narrow strip of resistive material. In my case, after 10 or 20 inrush current surges, this remaining strip opened due to overload, but not on all lots of the circuit - this caused some head scratching. I solved this issue by specifying surge proof (german: "impulsfest") resistors. They are trimmed differently - either a labyrinth cut or an L cut. Problem solved.

Seems to me like a possible fault in this case. If surge proof SMD resistors are not readily available, plan B is to replace them by Mini or Micro MELF - those can withstand much higher current surges due to being trimmed with a helix cut.

Hope that helps.

Greetings,

Rainer

[Mickle T.]

The location of the resistors in the circuits powered by four independent voltage stabilizers, low power dissipation at 1/10 of the maximum, as well as the absence of other faulty components, almost exclude the possibility of damage to the resistors from overvoltage, power surge, ESD etc.
Different multimeters (R6581, R6581T), different users, different operating conditions... The common thing in these cases is the huge operating time of multimeters (~100000 h), and (possibly) the same resistor supplier with a low MTBF due to embrittlement of the protective coating and the occurrence of microcracks in conductive elements. Later, I will check with our university materials science laboratory.

[dietert1]

R6581Ts were used in industrial setups, so it may have been in continuous operation for 20 years or more without any power cycle nor selftest nor calibration. Still it's hard to believe it was used accumulating failed resistors inside, so the failures probably happened after it was retired. So it seems like multiple old and brittle resistors broke by some mechanical stress or power on surge.
Will check those resistors in our R6581Ts. I remember carbon resistors would show symptoms before failure, maybe the same is true for thick film parts.

Regards, Dieter

[serg-el]

Diagnostic summary table R6581, R6581D, R6581T

[leighcorrigall]

I suppose there is a possibility that I installed the R234 backwards, but I was careful to check the other people's reference pictures before installation. Tonight, I will check the voltages.

EDIT: Couldn't wait!

R230: 98.5 mV, 0.985 V, 9.85 V
R234: 9.659 V, 7.090 V (across the network: 16.75 V)

So far, the voltage magnitudes appear to be reasonable to me. 

The firmware version states: ".A01". Is there a newer version available?

Without desoldering the R234, would there be any irreversible consequences if I tried to short the R234 to create a parallel resistance for the purpose of testing the diagnostic response? Maybe there is an easier method or maybe I am overthinking this problem.

I still believe that I replaced the original resistor network correctly. There are other examples on the PCB where parts were populated backwards. Some other R6581T owners have said that their diagnostic mode throws random numbers similar to mine. I wonder if this resistor is the reason why.

[leighcorrigall]

Diagnostic summary table R6581, R6581D, R6581T

FYI, my DMM is a T model that displays D during boot after the firmware update.

[Kleinstein]

If R234 would be installed backwords, this would cause the 17 V reference and other derived voltage to be way to low (more like 13 V) - that should be obvious and easy to measure. So I don't think this is the problem.

The difficulty is that it is not clear what the diagnostic values actually represent and how much math (cal constants) is actually applied.
Somehow the diagnostic values when using the slow amplifier all seem to be some 3 V too high at the ADC. This could be an HW fault, or maybe also a wrong cal constant for a rarely used offset (e.g. zero for the non AZ mode).

I think I have asked this before: did the meter read +10 V and +11 V OK (not saturating early - so not about a few ppm off, but more like 100 mV or more) ? If there is really this 3 V offset also there, this would drive the ADC in saturation and should cause an early saturation at the upper end.

A point to check may be of for some reasone the amplifer sometimes oscillates during the self tests. So maybe monitor the voltage at the ADCin test point during the diagnostic mode.

[leighcorrigall]

...

I think I have asked this before: did the meter read +10 V and +11 V OK (not saturating early - so not about a few ppm off, but more like 100 mV or more) ? If there is really this 3 V offset also there, this would drive the ADC in saturation and should cause an early saturation at the upper end.

A point to check may be of for some reason the amplifier sometimes oscillates during the self-tests. So maybe monitor the voltage at the ADC in test point during the diagnostic mode.

The instrument was calibrated and passed all test points with the exception of the high-ohms ranges. My R6581T measures the Fluke 732A 10 V source within a few ppm difference of a working 3458A. I do not recall if I checked 11 V, but I imagine it is fine as well.

The -1 V and +1 V measurements do not match well on the 10 V range. We are no closer to solving this problem either:

https://www.eevblog.com/forum/metrology/advantest-r6581-8-5-digit-dmm-mini-teardownrepair/msg3810377/#msg3810377