Free VE Table Corrections — Drop Your MSQ and a Datalog, I'll Do the Rest

[Nagrom]

ITT: Trust an anonymous stranger's vibe code to tune your car

 

[XCam34]

ITT: Trust an anonymous stranger's vibe code to tune your car

I get it but look at this https://turbobricks.com/index.php?t...a-datalog-ill-do-the-rest.383078/post-6584709 its accurate it works I understand you do you, if and when I get some one to send some actual logs then you will see it works...

 

[XCam34]

https://www.msextra.com/forums/viewtopic.php?t=82834 same guy?

yes

 

[HiSPL]

ITT: Trust an anonymous stranger's vibe code to tune your car

I'm open to it. AI is a powerful tool and its here already. Can't put it back in the box anymore.

AI does allow "normal folk" with an idea a path to completion. Just like everything else it could be good, bad, or indifferent. Just because AI was involved doesn't make it any better or worse than anything else.



Environmental impact aside that is.....

 

[XCam34]

Yeah man I use the sh!t to save time, I use talk to text and have it formulate a reply. Saves me time, but here i will use m finger tips only to communicate. If that makes everyone happy. Get me some F'kn real logs let me show ya'll this b!tch works, cause it does.

 

[740atl]

Looks like you're also on miataturbo.net

My question is this... How is your model taking user-tuned megasquirt data (and presumably other EMS's data) and fundamentally generating something meaningful out of it?

I mean can your model give a.... a refinement of existing data? Sure. I'll give you that maybe sight unseen... but is it generating something new... something more than what's already there? Maybe I'm fundamentally misunderstanding what you're doing/offering... That's a strong possibility.

The reason I say this is that almost no one who's doing MS is having their vehicles actually tuned on a dyno (why? well because it's expensive) The dyno is where the only meaningful MBT tuning can take place. With street 'tuning' we're guessing which direction the dart board is while blindfolded and then tossing the dart. We're taking an afr curve that someone somewhere recommended "was good" or "this works well". We're (most of us) not accounting for the odd lean cylinder. We're praying the shit doesn't detonate and defenestrate the block. I would like to be a more sophisticated user of my EMS's features but sometimes good enough is good enough. Does it run? Yes. Does it idle? Yes. Does it make boost? Yes. Good.

Aren't you assuming too much from the average MS or even aftermarket EMS user? Is this just glamorous algorithmically fueled GIGO?

 

[bobxyz]

Thanks for running my old tune & log through your analyzer. I compared the results from the default TunerStudio VE Analyze and your results - they're similar, but certainly not the same.

I'd need to dig into the log file to see why some of the weirder changes were made, e.g.:
VE Analyze changed the 90kPa row, around 2600 rpm from:
1800 2200 2600 3000 3500
68 67 68 73 71

To:
1800 2200 2600 3000 3500
66 65 _63_ 72 71

while your analyzer changed it to:
1800 2200 2600 3000 3500
67 65 66 73 71

Maybe 90kPa/3000rpm didn't get enough hits on box and the VE for the rest of the row is wrong (way too big)?


Other comments:
It looks like your rounding is wonky - for MS2/MicroSquirt, the VE table only allows integers so values like 82.4 and 79.1 get truncated to 82 and 79. I think MS3 allows at least 1 fractional digit, but why does your analyzer keep to mostly integers?

I'd like to see some additional graphs
- a graph showing hits-on-box for all samples
- a graph showing hits-on-box for just the qualified samples, or simplified to green/yellow coloring for well covered regions.
- a summary VE change graph showing just the non-zero deltas, maybe with the qualified sample region lightly shaded?

 

[XCam34]

Looks like you're also on miataturbo.net

My question is this... How is your model taking user-tuned megasquirt data (and presumably other EMS's data) and fundamentally generating something meaningful out of it?

I mean can your model give a.... a refinement of existing data? Sure. I'll give you that maybe sight unseen... but is it generating something new... something more than what's already there? Maybe I'm fundamentally misunderstanding what you're doing/offering... That's a strong possibility.

The reason I say this is that almost no one who's doing MS is having their vehicles actually tuned on a dyno (why? well because it's expensive) The dyno is where the only meaningful MBT tuning can take place. With street 'tuning' we're guessing which direction the dart board is while blindfolded and then tossing the dart. We're taking an afr curve that someone somewhere recommended "was good" or "this works well". We're (most of us) not accounting for the odd lean cylinder. We're praying the shit doesn't detonate and defenestrate the block. I would like to be a more sophisticated user of my EMS's features but sometimes good enough is good enough. Does it run? Yes. Does it idle? Yes. Does it make boost? Yes. Good.

Aren't you assuming too much from the average MS or even aftermarket EMS user? Is this just glamorous algorithmically fueled GIGO?

So to clarify what I'm actually doing: VE table correction, only No timing changes, nothing else. MBT/timing is a completely different beast you're right that you can't do that reliably without a dyno...


On the GIGO that's honestly the part I've sweated the most. There's a lot of filtering happening before any data touches the table. Post-AE cooldowns, post-fuel-cut cooldowns, off-throttle non-cruise gating, neutral-AFR rejection, wideband delay correction all of that exists because raw logs are full of cells that look like they need a correction but are really just transient artifacts. Feed it a log of someone bouncing off the limiter with a half-dead O2 and it'll basically refuse to touch the table, which is exactly what it should do.


So that's what it is: a more careful pass at the same data TunerStudio's already working with No magic, not a dyno replacement. It's aimed at people already running iterative log-and-tune cycles. If that's not you, totally fair not everyone needs it.

The whole reason I am building this thing is cause of what you are saying the tuner studio software really is shooting in the dark , I really wanna tell you more about what the tool does but I feel like I'd be giving away too much info. But... The filters it has a lot to do with that, I feel like tuner studio is averaging in fuel cuts accell enrichment widebang lag etc, my tool does pretty darn good at reading and presenting steady state fueling for averaging....

 

[XCam34]

It looks like your rounding is wonky - for MS2/MicroSquirt, the VE table only allows integers so values like 82.4 and 79.1 get truncated to 82 and 79. I think MS3 allows at least 1 fractional digit, but why does your analyzer keep to mostly integers?

Because I am dealing with tables like this, and my tool is a designed to be a multi pass tool, I am confident enough to try these fixes on strangers cars I am not confident enough to bump a cell more 10 ve points like tuner studio will... I'd rather make multiple small passes rather than throw the kitchen sink at it,. On integer tables it's intentional decimals truncate anyway. On 0.5-resolution tables I'm being conservative on purpose because I'd rather make small whole-number passes than chase noise to a tenth

1500×20 (111 frames): TPS = -0.70 across all 111 frames. Not 0.5, not 0.2  literally the same value every frame. That's the throttle blade physically against its stop, TPS reading slightly under its zero-point because of cal offset. RPMdot median -30, min -120. DC at ~3% (minimum). PW at 1.2ms (essentially injector dead time itself). Engine is decelerating with no fuel commanded.


1800×20 (46 frames): TPS = -0.70 again, all 46 frames. Same picture. RPMdot down to -130.


2200×20: Zero raw frames in the log. So VE Analyze pulling -4 there is literally a smoothing/interpolation hallucination  it's pulling from neighbors that are themselves contaminated.

 

[bobxyz]

Thanks, those new graphs are a big help understanding what regions were well covered by the log, and what cells were changed. For the final graph, I'd try using a green background for cells that weren't adjusted (OK as is), instead of the dashed green border.

 

[XCam34]

Alright now I just need some real logs....

 

[XCam34]

Anyone ?