About a month ago, I purchased a broken Keithley 199 multimeter off ebay for about $70 including shipping. From everything I’d read, it was a pretty accurate 5.5 digit, 310,000 count multimeter with a nice bright LED display (as opposed to the Fluke VFD, or HP non-backlit LCD displays). Having bought sight unseen, I wasn’t sure what exactly was broken. It could just be simple user error, or something more serious, but from the age of the unit, it should be fairly repairable.
And thankfully, the manual supplied by Keithley (available here, via KO4BB, without registering), has wonderful block diagrams, layout images, and schematics. After receiving the unit, I found it was indeed non-functional, with wildly varying readings. All the buttons worked, and it would switch ranges, but the readings never worked.
Working with some help via the eevblog forums, the general thought was to confirm the input path using the schematics and a highlighter to mark where the signal was found, etc. Thankfully the Keithley 199 has a diagnostic mode, which allows you to look at the the various inputs to the multiplexer. The meter normally, several times a second, cycles between 0V, 0V, -2.8V, and the signal actually input. It uses these to drive an integrator that looks at the rise/fall time to compare those voltages. Read the “theory of operation” for the Keithley 199 for more info.
Anyway, The signal made it all the way to the integrator, and then disappeared (which I thought was an issue, but wasn’t). I started looking down that path. But when put in the proper diagnostic mode, I could see the integrator working. So that wasn’t it. I had already checked the voltage references in the unit, and they all checked out. Obviously so did the analog ±15V, +5V, and digital 5V rails. What didn’t seem right was the clocking on the analog side of things. So I started looking down that road.
Interestingly, in the start of things, someone on the forums suggested looking at the A/D counts on each side of the opto-couplers. So looking at those, things looked fine coming into the analog side. Coming OUT of the analog side looked bad. So back-tracking, everything went through a hex-inverter (U49 according to the schematics, a 74HCT04). On that, I was getting kind of crap output on the clock signal going further into the analog section, and the A/D counts heading out to the digital side of things were being corrupted as well. So, after ordering a replacement, I unsoldered that IC and tested it on a breadboard, and it worked (after wiring correctly). I installed a 14 pin socket on the board as well, so I could install a new chip when it came. Huh. So I put it back into the board (in the socket), and the meter started working properly. Well crap. I found the issue, but the IC that didn’t work, now seems to work. Talking it over in the forums it seems likely the heat, and mechanical stress of unsoldering probably made a contact a bit better, but it would likely fail again. So, once the new chip arrived, and replaced it, and it still works. =)
So, there are a couple learning experiences here, since this was by far the most complicated repair I’ve done by far. 1. Check your voltage rails. 2. Check your signal path. 3. Check your glue logic. You might be able to swap 2 and 3, depending on what the symptoms are, but I think I probably wasted a bit too much time looking at signal path issues, and should have trusted what I was seeing earlier when I first looked at the glue logic and saw weirdness.
After all that, someone else on the forums found similar failure in that hex inverter. So either there was a batch issue back in the 80’s, or where-ever these meters came from exposed them to something (drop, ECD, etc) that toasted that chip prior to anything else (which is lucky for me). On a big note, thank you test equipment manufacturers for including schematics with your manuals. It’s a bit sad you don’t see it anymore unless you find the service manual.