Having stuck my toe into the waters of getting an engine conversion to run on aftermarket EFI technology, and more specifically HP Tuner for LS swaps, I have come to the conclusion that any number of people can get a swapped engine to run, but very few run very well.
So how do I get a chatbot to read my data-logs and produce a tune for a street driven swap? And why has nobody started selling suck a service?
Pete
You're looking at the wrong sort of AI; chatbots are horrible at math. There are several purpose built algorithms to tune fuel tables based on logs of wideband O2 data.
A spark table would be harder. There's no feedback. You might be able to get a neural network to take a database of good spark tables matched to engine mods and have it predict spark tables for other combinations. The trouble is getting the right data to train with - you don't want to scrape it from forums as you'd get a lot of really bad tables mixed in.
In reply to MadScientistMatt :
Your last sentence is what made me do this post. The state of amateur tuning is much akin to those infinite number of monkeys banging on typewriters in hopes of doing a Shakespeare play.
I also doubt that any solution would work without a dyno to create repetitive loads.
Most of what we call AI, including large language models, are statistical analysis tools. And they're no better than the statistics you give them. Many OE ECUs these days are what are called model based, where they collect a massive amount of dyno testing data and have a computer fit a set of equations to their data. But that type of analysis requires more dyno time than conventional tuning, not less. You only see it on OE application and some very high end pro racing events.
I could see a dyno shop with a huge library of dyno tunes making approximate tunes to order by averaging or blending tubes, but this would just be an automated way of doing what experienced tuners may do manually to make a starting point for a tune.
I could also see the scammers making bogus tuning devices (originally resistors inline with the IAT sensor, more recently LEDs that plug into the OBD2 port) claiming they're doing this, if they aren't already. And I have no doubt chatbots are good at claiming a terrible tune is great since they have no concept of reality underlying the data.
But without good data, either from your engine or a massive library of good dyno tunes, AI won't be much help.
I suspect an adversarial network approach could eventually generate some pretty good stuff. Might not make economic sense for someone to do, though.
DIYautotune.com is pretty close to what you're asking
Are you talking aftermarket computers or OE computers? HPT tunes OE computers.
Tools already exist to get a good tune with HPT. Some of them are pretty basic, like take a histogram of X behavior and multiply a table by the difference, table smoothing, etc. Good tuners write their own logging parameters and math channels to do the dirty work.
Really, 99% is stuff you should not be touching. Your main focus is on VE, MAF, and idle flow characteristics. The rest is all lofty stuff like intake manifold wall wetting that doesn't change, or really esoteric fuel injector characterization that is easiest dealt with by using OE injectors (aftermarket are 90% junk) and copying the tables from whatever those injectors came with.
HPT used to have a really good tune repository. Got a set of injectors from a (blank)? Copy its injector tables.
LS swaps are really easy because GM made the computers kind of self contained, and so you just use that. No tuning needed other than disabling VATS, recalibrating the speedo, maybe fix the MAF transfer curve if you are not using the original air inlet and airbox.
In reply to MadScientistMatt :
GM ECMs very much are of the type where they model the engine characteristics first, and then there are separate tables for "tuning". They have been this way at least back to the mid 1980s.
If you want the engine to run a little leaner at WOT and 4500rpm, you don't tweak the VE or MAF, you tweak the WOT enrichment curve. Assuming that the actual fueling is the same as what is intended.
In reply to Pete. (l33t FS) :
You make it sound easy and I guess it is to anyone who really knows what they are doing.
What I have observes is a few attempts at doing LS swaps with either the GM ECM or the Holley Terminator system and a "Sloppy" cam. So far I am not impressed with any of the results. They go really fast when you stand on the pedal, but horrible if you had to drive in Atlanta traffic during rush hour.
I did use the word "Chatbot" however I do realize that that would be the wrong term. I just have a feeling that this game could be a lot easier if the random variable of the inexperienced amateur car builder could be removed from the equation. Its an air pump that sucks in explosive air and makes a big boom based on measured parameters.
In reply to NOHOME :
Drivability is what most canned tunes and even a lot of " so called" dyno shops miss the mark on. They nail the WOT stuff, but thats the easy part. Cold starts. On/off throttle transitions, part throttle cruising under various loads, etc. Stuff all takes a large amount of time that most people in general don't want to pay for.
I just did a blown big block with a procharger and a holley fuel injection system. I spent about 10 hours doing not just WOT stuff but a ton of drivability and part throttle stuff, as well as setting up the electronic trans, lock up points in the converter, shift pressures, etc. Makes a HUGE difference in the usability of the car. But 10+ hours at $150/hr.. isn't cheap. But that car is now an 850hp Bob Costascat around town, cruising down the highway, etc... until you nail the throttle. In which case it'll do rolling burnouts at 70.
There is a lot of material out there you can read, but a lot of it is also just plain incorrect or wrong. Also the "same" combo can react very differently in different vehicles. You'd be surprised. I've set a few gen v swaps up, and even from car to car with the same basic modifications...they have tended to like slightly different fueling strategies.
Being said, the tuning school books I found to be pretty decent. Theyre pricey...but some good info specifically for HPTuners and they have a course on LS stuff as well as another one dedicated for Gen V DI stuff.
And check out Goat Rope Garage on the tubeofyou. He does some fairly decent videos on drivability stuff for LS'S and HPTuners. Like I said, there's lots of good info out there. But it's buried underneath all the E36 M3.
You'll need to log in to post.
