Using AI to analyse tuning logs

[Charlescrown]

Just started using Google AI to analyse my logs and it's really been a great help. The car now has a heap more power which I thought was a mechanical issue holding it back. There has been occasions where it made errors but that was a communication caused by me not providing sufficient information but when questioned it realised the error and we got back on track. Hopefully the 2 of us can end up with a successful tune.

[antus]

Good to hear. I have been using claude lately and also getting good results. Did you give it the bin, xdf, adx, and xdl files and let it go from there or give it some copy pasted stuff?

[Charlescrown]

Sure did and it is great with 12P info so I will keep going. It references many site data which is more than I could and the analysis is very informative and to me accurate.

[antus]

I wonder how much info it gathers from its users as well. I have given claude access to the 12p source code, so I do wonder if it takes learnings from that for other people. I have it set to not use my data, but I'm not sure I trust it. That did help it to find an issue with my tune that affected mixtures at the exact switchover from cold engine to hot at a certain number of degrees (21 from memory). I could see the degree scaler change at the exact moment mixtures went off and it had mis-diagnosed it the first time so I drew its attention to the exact temperature change and it went to the source and confirmed the behavior and fixed it in the calibration properly.

[onslaught]

I've had it analyse datalogs and make suggestions that were way off. It's a useful tool but while it speeds up research, it can't replace knowledge.

[vlad01]

That's really interesting use case. I never would have thought of that.

I have been using it to design an exhaust system based on the Vizard info, and it gave some interesting insights. I couldn't find much head flow data out there, so I told it what the intake flows at 25" at 0.500 since I know that figure for the hypothetical engine I have in mind. But since I never had the exhaust flow, I used it to infer what it might be based on stock figures and potential based on what a good head porter would do. It pointed out that the Buick has very atypical heads from factory that have disproportionally good exhaust port flow or sub par intake compared to the ex if you like, and that the in to ex flow ratio is way bigger than nearly all engines out there. It said that a good porter will be a little bit more conservative on the ex to improve the ratio by lowering it to the ideal 70-75%. That then allowed it to guess the flow figure of 190 ish CFM at 28. That then allowed the header deign to take place. Of course I will have to ring up HSD and nag them for the figures they never got me all those years ago, because it might be off. But I though it was neat and I learned a lot. Due to this, it said I should use a single pattern cam which I happen to already have, so that the exhaust doesn't over scavenge at the lower to mid rpm based on the extremely good ex port compared to the intake. All this matches what I have reverse engineered from photos of race engine in question using known scales within the pics. The AI and I determined the engine's headers were made on the acoustic wave model rather than the velocity+flow models Vizard is famous for. That was basically my aim, to change the system to work better for what I want and get the most out it in my use case.

My goal was to get a baseline for a system that is favored for best area under curve as opposed to peak power only, so the race engine suits hot street use instead.

I even got a figure for the collector length range that should work which saves me having to buy the book lol.

Most of the other aspects of the system is said I already had the best plan, so good to see I am getting my head around it before I ever asked AI.