...
First off, Pat, thank you for your detailed response!
Now: as to the Fluke 289 and its logging, it also counts "events." I'm not an expert on this complex instrument but I those "events" are things that happen below the resolution of the capture interval. It shows a count of these things, but to see the details you have to download the data to a computer (which I didn't do). However, I did not notice the "event count" increase, so maybe it still didn't see anything. I have a Rigol DS1000E so I could also scope it. (I did that in the past to measure the stability of the power supply; it doesn't have more than 80 mV of ripple on the 5V or 12V lines at around 10-20 ms period. Pretty noisy looking on the scope until you use the averaging feature.) Since the computer starts acting weird after 15-20 minutes I doubt I could record enough data in enough resolution to see anything, anyway, without having hugely fast reflexes (and it's super boring to watch something to see when it crashes).
All that being said, I wouldn't do anything to the caps unless I was quite certain they were bad, and I don't think you're there just yet.
I welcome thoughts as to what else to do. Here are my thoughts:
1. Next time it crashes, quickly disassemble the computer and look at it at with a FLIRone camera. This essentially entails pulling off a backplane connector, pulling out two VMEbus-style boards, and examining them and the internal backplane. They will probably cool down quickly during this so I am not sure it will be of any value.
2. Replace or recap the power supply, see if it helps. It has 4 feeds: 5V, 12V, -12V and -5.2V (no idea what that last one is for).
3. Pull all socketed chips, spray contact cleaner in the chips and sockets, and reseat. There are probably 60+ normal sized DIP chips, plus maybe a dozen large DIP chips, three huge square chips (one being a Weitek FPU) and one that is a chip with pins on the side that goes into a square slot with pins on the side that I have no idea how to remove. I fear doing this because I haven't removed chips in decades and breaking any of them spells the permanent end of this irreplaceable computer.
4. Measure current on the 5V and 12V rails. I believe this power supply puts out 35A on 5V though, so I'd need quite some big connectors for the Fluke. (Todd MAX-354-1252) A watt meter connected to the 110V line shows it drawing about 230W in stable state operation. See how stable that current is. I'd have to rig something up to get the current through the meter, or get one of those probes that can go around the wire (but those are ridiculously expensive when I last looked).
Other ideas?
He's referring to how a cap behaves when connected to an ohmmeter.
I understand how caps behave in general; my question was more about how to use the resistance function of the meter to discern whether a cap may be bad in-circuit. I don't know what sort of voltage or current the Fluke 289 puts out when measuring resistance, but I can't imagine it's much. For the caps that measure in the kilohms, you definitely see the resistance change while measuring them. For the ones with 5 and 8 ohms, the resistances stay pretty stable.
Frankly, I don't know how to test the caps in-circuit or if there is even a reasonable way to do it. If these are bypass caps as I expect they are, then measuring one is probably like measuring them all (and makes me wonder why there would be different resistances at all in the first place).
...instructions on replacing the caps...
So, it sounds to me that replacing the caps on a 4 layer PCB is no different than doing it on a regular PCB, from your general description. I was concerned that there might be connections, somehow, to the internal layers. The solder exists on both the top and bottom of the board for each cap, either by design or pulled through by capillary action.
Thanks!
Doug