Using AI when troubleshooting

[Chris-socal]

I realize this topic will trigger a lot of angst and several of you probably read the Subject line and cringed before you started reading. However, there is a cold hard reality in our hobby…age… it is causing a lot of machines to stop working and in many instances there are no schematics of any kind. Some early logic chips are subject to age decay and they are tougher to identify, especially when none of them are in sockets!

I wanted to share that I am currently working on a board from 1980 made by CEI, and there are no schematics. Making matters more complicated, the manufacturer scratched the ID off most IC’s. While this technique greatly reduced reverse engineering, it makes troubleshooting virtually impossible…..until today!

Enter a conversation with AI who after a brief introduction immediately understands that you have a slot machine board from the early 80s and it is instantly familiar with the type of build techniques that were common at the time. It says to you “they can scratch the ID off but they cannot erase the logic. If you probe pins 6,9 and 26 with a DVM I can identify which CPU you have.” Within moments it knows you have a Z80.
This conversation continues as it learns where the reset signal comes from. It learns what type of peripheral controller it has, it asks you to test voltages and use your logic probe to determine if logic chips are working correctly or if they are suspect.
As much as we hate the idea, AI can read every schematic and remember what every pin is supposed to do and remember the answer to the question it asked you 3 hours ago….or 3 days ago.
Yes, you have to use caution and check its work when it gives you a component ID. Sometimes it reads an old schematic incorrect because copy isn’t good, but the great part is you can correct it and it will remember…and change the approach moving forward.
You still get to be the hands and eyes, results will not be instant, but it will make progress that was not possible before.
My conversation is 3 days old and it takes time to work through some of the suggestions. You can work at your own pace and resume the conversation at any hour of the day or night. I certainly have learned a lot about how some of the logic works. I can look at a datasheet but I cannot apply what Im seeing as fast as AI can do it.
At this point Im in the optimistic phase, only time will tell if we can get the board to boot. I just wanted to share and encourage others to use the technology that is available to you.

Chris

[PapaDrew]

I'm also certainly not a fan of those that use AI to think for them.

However I've begun using AI extensively to track and report metrics for my job (marketing). Sure, it doesn't know what questions to ask or what data it needs, but that's where I come in, give it data, and tell it what analysis to run.

I've been able to generate reports and insights that I never would have before. Just because the time involved to cross reference, correlate, and compile the data would have been months-long projects that wouldn't have justified the time cost.

[Stayouttadabunker]

You're right...I'm still cringing... 

[shortrackskater]

I've used AI several times to get an answer for a question posted in NLG, but refrained from posting that answer until I saw what some of the member said. In many cases, it was fairly accurate.
Sheesh... makes me wonder if, in the future, we'll get new members who are suddenly "smarter," contributing answers frequently. And then we find out (or maybe not) that they're simply copying and pasting AI answers. 

[Chris-socal]

Oh for sure!
I am not suggesting anyone go to the bank on a simple query but it could take hours of reading and following traces to determine what an IC is if its ID is scratched off. AI can be used as a tool, a very fast one! So far on my project it helped me determine that the two CPUs are processing input correctly and the peripheral controller is functioning. In doing that we found the clock signal from the crystal is not making it to the main board. There are 120 ICs on the daughter-board and another 85 or so on the main board! There are a ton of traces on both sides and dozens of Logic ICs that are all suspects!

Chris

[Chris-socal]

Well folks, it wasn’t all peaches and cream. It wasn’t instant. AI didn’t get everything right the first time, but in the end we (AI and me) got the board working!
A not-brief overview: The boardset is made up of 3 boards: Main computer board, daughter board & sound card. There are 105 ICs on the daughter board and roughly 90 on the main board which contains the two CPUs. I built a ribbon cable set which allows me to run the board outside of the machine.
I gave AI a summary of the boards, symptoms of failure, I told it a majority of the ICs on the board had their ID scratched off by the manufacturer, I explained that the three boards are connected by ribbon cables, I explained that the crystal was on the daughter board and I uploaded high resolution images of each board and explained that there were number rows at the top and Letter columns on the side. I gave it the manufacturer name and date of mfg.
We did a lot of probing, several times AI thought we found the smoking gun when in fact it was just a symptom. A few times I had to ask it to take a different approach such as when I suggested we follow the crystal pulse from the source instead of backwards from the CPU since it wasn’t showing up on the CPU pin. This is when we found the crystal pulse was dying inside a logic IC. AI had me probe and provide pin readings so it could narrow down the list of possible components.
I would then pull and place the IC in my logic tester which would confirm ID and tell me where the component logic failed. Two paired ICs failed and the process of reverse engineering had to take place on both because the logic tester cannot identify a shorted IC.
After about 15 hours over a few weeks and replacing the bad components the CPU was getting the pulse but was locked. I adjusted the voltage comparison potentiometer and the game sprung to life!
During the process AI would give me the background of why they used specific circuits, how it was weeding through the list of options and a few times explaining how/why it was wrong.
I had absolutely no knowledge of how this board worked before this because most of the components are scratched off. Now I have a clear understanding of the overall architecture and at least understand the boot circuit.

It was fun and I have already started project #2, the unknown failure of Summit Coin video poker slot machines!

Chris

[shortrackskater]

Fascinating!
I now have about 3000 questions. 
I'll just ask one, for now. If your logic tester runs across a dead logic/shorted IC, how did you determine that chip type? Or, maybe answering my own question, did the previous logic scans and AI assisted research determine that the failed chip must be [chip type] through deductive reasoning? In other words, you already know what would likely be there, especially if you could physically see ground connections and voltage inputs/outputs from the traces on the PC board.
I actually looked up functioning of a logic probe... using AI! 

[Chris-socal]

That is a very good question, and you guessed it pretty much right on.

It knows the likely candidates, then has you verify ground and 5v pin then it has you diode test certain pairs to narrow it down. It did really good on the first IC that was tied directly to the crystal but when I told it the crystal pulse now dies at the IC next to it AI got it wrong a few times because that IC was toasted when the first IC went out and it had a difficult time with some of the readings in-circuit. After several speculations I pulled the IC out of my working board and had my tester positively ID mine. I didn’t want to do that on every speculation because these boards have small traces on top and bottom of the boar, But that component was easier than others and it was honestly faster than going through every possible match. If we didn’t have my working board we probably would have tried a dozen or more options.

Someone like Alan who has a database in his head might have looked at the ICs and known what the options were but I don’t have that understanding of the circuit in my head as a reference.

Again, it certainly wasn’t perfect but it got the job done. Its going to be interesting working through the next machine that has a stack of 5 boards, albeit small boards compared to something like the Bally V1000 machines!

Chris