Pontiac G8 Forum banner
1 - 3 of 3 Posts

·
Registered
Joined
·
237 Posts
Discussion Starter · #1 ·
I have always wanted to know what is meant by the ECM “learning” your driving habits. No one has ever given me an answer other than speculation, and I find no literature on the subject anywhere. Speculation? Well, I can do that myself. Here’s what I think is meant by ECM “learning”. Comments are welcome.

The LTFT’s are determined by integration and temporal filtering of the STFT’s, with the STFT as the feedback, and the LTFT as a pseudo-feedforward. STFT’s are instantaneous – they adapt for given current readouts on a 1 to 10Hz timeline (limit cycle carrier) depending upon the car, 02 sensor slew rate and ECM gains. LTFT is actually a “vector quantity”; the LTFT integrals are a function of MAF / fuel flow. FT’s are required because of calibration errors in the MAF and injectors, as well as imperfections in the assumption that all fuel gets mixed, assumptions about VE (i.e. not all the air that the MAF sensor passes actually gets used due to valve overlap, etc… but let’s ignore that temporarily for the purposes of discussion). And those cal errors are different for different flow conditions.

The MAF and injector calibration errors are a function of the flow rate (i.e. there’s some degree of nonlinearity, not just a simple scale factor error and bias). So it stands to reason that the LTFT algorithm would acknowledge that dependence in some way. The most obvious way would be to model the dependence with some basis functions of MAF and just observe coefficients, but that’s probably not stable and too simpleminded anyway. But what is actually done is to divide the MAF / flow into “bins” – about 6 of them, let’s say, and they run a separate integral for each bin. The bins are something like “idle”, “moderate accel”, “decel”, “heavy accel”, etc, etc, and note that the bins take into account more than just MAF. But that’s just a nit – the point is that there are these “bins” and they correspond to different flow regimes as sensed by the MAF.

So what’s the “learning”? LTFT’s get reset when the car starts up, and they are set to some default that’s fairly conservative – meaning “rich” and not optimal. The STFT’s will clean things up and adjust the LTFT integrals on a by-bin basis from there as flow observations for each bin occur. But STFT’s can’t respond instantly. The limit cycle might only by 2hz, in which case the adjustment would be so slow that you’ll definitely see transients in real AFR that might be rich or lean, which will reduce power either way (either through cold mixture or pulled timing).

So what if you have not yet driven (since the last key cycle) under conditions that meet some of the bin criteria, such as if you have not driven at WOT for a month? Then the LTFT’s for those bins are at their defaults and when you first go into one of those bins, the tune will be suboptimal since the STFT’s can’t respond instantly. It takes some time in that bin to get the LTFT’s to adjust out the cal error for that bin. But once you do that, the next time you go into that bin, the LTFT for that bin is good to go and the STFT has less work to do. You won’t see timing getting pulled or other artifacts that happen when the STFT is futzing around getting the cal adjusted for that bin.

I think this is what is meant by “learning”. If you go thrash the car around in the parking lot before you start your drive home, all the LTFT’s will have an up-to-date adjustment stored and you’ll probably notice a little smoother / better response when you step on it. Although I doubt that this is a good strategy since the car will not be warmed up and the LTFT’s you get in this situation might be a little off. Temperature is probably a huge factor in the validity of any of these calibrations.
 

·
Registered
Joined
·
1,875 Posts
So what’s the “learning”? LTFT’s get reset when the car starts up
This is where I see an error. LTFTs are not reset when the car starts up. You have to either command a clearing of the fuel trim cells through software, or the ECM losing its 12V power signals that the ECM may have moved to a different vehicle and needs to relearn the new calibration. There would be no benefit for the LTFTs resetting at each startup since, in general, the configuration of the vehicle has not changed, and any small variable, such as weather changes, would eventually be accomodated through STFTs transferring to the LTFTs.
 

·
HPT Pro Tuner
Joined
·
3,815 Posts
+1, plus some other interesting things to note, STFT's are actually instantaneous in their response, LTFT's are nothing but an average over the long term of what the STFT's are seeing, and mostly used to tell if the ECM is actually maintaining proper fuel control over time.

Here is a little more explain in general.
http://www.stealth316.com/misc/obdii_fuel_trim.pdf

I run my car with STFT's only (LTFT's disabled in my tune), and they keep my fueling where it should be while in close loop. I can also see and modify where the % of STFT's might be needed if i do any modification to the engine, instead of letting the LTFT's do this by aproximation. they're needed on a bone stock car, not needed when you can change things around, especially when you have a cam, or maybe i'm just lazy and i don't want to play with LTFT's boundaries :)
 
1 - 3 of 3 Posts
Top