Neural Network discussion

So as an update to the ANN and WOT fueling ... I finally went in and retrained the ANN to address the EQ leaning out at the top of the RPM range during WOT. Here is the result:

Attachment 158251


Attachment 158252

what else did you change in the tune???

lol

and how did you get it to go so fat on tip in when at cruise??? (see pic)

i am still trying to figure out that out...

post a log of it leaving dead stop/dead hook and running thru some gears, not cruising and you rolling into it in stages, and one gear only

thanks

what is going on with the rear cylinders???

I think your scanner config needs work ... it's not fat at tip in.

Attachment 158258

Commanded EQ lags, so I'm using a math channel which shows "Stoich + FA Enrich" for determining commanded EQ.


Here's the compare log between the two tunes. I have made various other changes, but none fuel related except for raising the COT/EOT threshold. It's the airflow change (ie: The ANN training) which made the difference in the WOT FAR pattern

Attachment 158259

The timing changes are just me scaling out those tables for boost, in preparation for my next phase of this journey. I'm still only hitting 10-11 or so max spark, I just scaled axis of the tables.

was referring to actual...

not commanded.

i think it being fat there is the difference between the GPEC2A and the dash 18

I guess I don't know what you mean by "fat" then. I interpret that to mean "more fuel than I want" ... but it looks to be hitting my target pretty well.

0.76 is pretty close to the 0.77 commanded. It's just your screen shot shows the commanded at .920 because the channel you are using to show commanded is delayed.

So the -16.8% EQ error you're showing isn't accurate. Its really -3% at that moment in the log.