I've reported the failure of these instruments in high Ohm ranges (1MOhm and upwards):
https://www.eevblog.com/forum/blog/eevblog-544-fluke-5450a-resistance-calibrator-teardown/msg1113471/#msg1113471PeLuLe proposed replacement MEDER reed relays HI05-1A66 with 10^14 Ohm isolation, which were not available until recently at DigiKey, for about 9€/EA. (PeLuLe, thank you very much)
The old relays really had >20pA leakage current @ 31V, whereas the MEDER relays were < 1pA, probably 100fA only, but hardly measureable with my 34465A.
I pulled out K16 .. K22, K30 and K31. This already gave a big improvement on 3458A measurements, directly inside the resistor chain, as already described in the foregoing thread.
I did not want to screw up my instrument completely, so I removed these relays, plus sockets of K30 and K31, as these were also specially assembled with PTFE wires.
On the bottom side of the analogue PCB, there's a 15mm clearance between PCB and metal shield, so enough space to assemble the 8mm wide reed relays here.
I soldered these nine relays over the pads of the reed relay sockets, without generating contac.
Doesn't look nice, but fulfills its purpose, and can easily be restored to original state.
I used very few additional solder, but instead the residue from the existing solder joints.
So I had to remove very few flux afterwards, using very few methyl alcohol.
The result is very good. The systematic decrease of resistor reading vanished, so the verification of the freshly calibrated 5450A is now consistent with the 3458A and 34465A reading.. attention here,.. inside the 5450A specification limits ( which was not the case before), and well accepting, that the HP instruments do not have good enough high Ohm measurement capabilities (10 times worse than the 5450A spec.)
Disadvantages of these relays are higher resistance (about 50mOhm / contact), and a higher thermal voltage (up to 50 µV), compared to the original relays.
These errors can be corrected by usage of 4W and OCOMP, in contrast to the leakage currents of the defective relays.
For the 1MOhm to 100MOhm ranges, to compensate the e.m.f. voltages, I therefore had to make use of either DUT reversal method, or to directly measure the e.m.f. voltage and adding/subtracting the corresponding resistance values.
Anyhow, I now trust the 5450A much more, and the big uncertainty of the 34465A and to a lesser degree of the 3458A is now much more tangible.
Especially the 100MOhm range is really mediocre, on the 34465A you can observe big, big variations, depending on temperature, for example.
Before this repair, it was not clearly to be assigned to either the DMM or the 5450A.
At the moment, I really could not calculate or determine the achieved uncertainty of the ranges of the 5450A, maybe better than its specification.
So either a strict error calculation, or a better resistance measurement method is required / welcome.
Anyhow, it will now serve well as a calibration standard for 6 1/2 digit DMMs, as it for sure fulfills again its specifications, also compared to the similar modern 57xx calibrators.
Frank
For those, who want to use that method, I append my calculation sheet.