Been using HP tuners to log STFTs and then adjust the VE table to achieve trims between -1 and 1. Everywhere I see mentions that having trims near 0 are best for driveability and fuel mileage etc.

Now what I don't get is why? I mean no matter what is happening with the car, the job of (at least in closed loop like I am running) the o2 sensors is to keep a constant stoich AFR of 14.7. If I'm dumping too much fuel, they take it away...if it's running lean they dump fuel, all to get it to that magical 14.7 AFR that the narrowband sensors can read.

So regardless if I am running average trims of -15, 0 or +15 the end result is the same 14.7 AFR. What is the point of telling the PCM the exact fueling to run so that it doesn't have to add or take away any fuel when if I tell the PCM to dump X amount of fuel and it sees it as too rich it corrects that and takes away fuel from what is commanded and runs whatever fueling it needs for 14.7?

Am I understanding this wrong? Is there a difference in the amount of fuel dumped in the car at -15 vs +15 if it always keeps it at the same stoich AFR?

 

The STFT and the LTFT when added together should at a small negative number. When you go wide open throttle the trims will lock a zero, then you can set the PE by reading the WB. The trims do what you said under Closed Loop. Open loop is what you are tiring to fix. Besides a well tuned VE table will give you better transition fueling.

 

Heh, you probably don't even realize how good this question is

There's two main ways of controlling things: estimation done by creating a physics based model (i.e. Speed Density), or for systems for which the physics gets too complicated, you got purely reactive, you got 'controllers' constantly monitoring and adjusting things.

There's pros and cons to both approaches. Physics based models tend to be much more precise, but don't work well in environments for which they were not intended for, or where you introduced a new variable (they don't work when sensors aren't up to their operating temperature, or hardware is not sealing and the air is leaking). They tend to also be complicated (think Manifold Temperature estimation in the GM ECUs), as they try to make them be more robust. Some of them are actually computationally intensive, which restricts their applicability.

On the other hand, reactive 'controllers' don't give a **** about physics or correctness of what they're actually adjusting. They're just there to adjust things as needed. Most of such controllers are a version of what's know as the Proportional-Integral-Derivative controller, or PID for short. They are a mathematical way of being able to follow a signal--ANY signal. That's the cool part, as you don't have to bother with building and validating a physical model, so for a system where you got a lot of variables and conditions (like AFR of engine) you just set it up, and let it run. So why don't just just run everything entirely on PID's? PID's must be 'tuned'. Every system has their own characteristics, and you must dial in each controller for that particular system. So you're back to doing calibrations, which was an important reason to get away from physics-based models. Another problem with PIDs is that PIDs are purely corrective, thus they don't really know how to start. Math people would call it being 'sensitive to initial conditions.' So how do you start your engine if you're running purely on PIDs? It has no clue how much fuel it wants to start, but once you get it going it can figure out how much and which way is it off, and make (hopefully) a good correction.

Another problem with these two approaches is that they're really not comparable. Estimators work from prior knowledge, and physical properties. Controllers work 'on the other end' of the engine, they observe and correct, but don't ever expect them to come up with their own estimation.

So just like with the MAF/SD hybrid system, GM went for a hybrid approach too: ECU makes some estimations first with MAF or SD (or both, or none if it's using lookups in case of a cold enigne, or cranking). Then the ECU commands a pulsewidth based on that estimate. Things go boom, exhaust fumes go around the O2 sensors, and now we make an observation, an estimate of quality of the estimation we've done just before the explosion. Now we can make an educated correction on an already decent estimation. This way it's not purely reactive, so you don't have any problems starting things. It's not purely predictive, so it's flexible enough to learn about current conditions (short trims), or long term wear and tear (long trims).

Then there's useful aspects of the hybrid approach that would have not happened by using either subsystem solo; In a purely reactive system, there is no such thing as 'too much correction' so if there's an intake leak and your correction always says 'need less fuel/too rich', you will ultimately cut off all fuel to the engine, which is preposterous.
Now thanks to the airmass estimation we know what the airflow should be, while allowing some slack/slop in measurements. So while 10% correction isn't great, it's not the end of the world. However, if the trims are more than 25% off either way, the ECU let's you know that something went wrong, because that's outside of the sane values for this calibration. Purely predictive system would just keep commanding what it would always command, ignoring any changes in the system, whether they're environmental, or your hardware just borked itself.

So why tune VE? Because it's the sane thing to do. Let the ECU run the way it was intended. If your tune is perfect, your corrective mechanism will be very quiet, because the estimation is already doing a great job. If there are discrepancies/noise in the system, the corrective measures will be taken, but within reasonable limits, and you can also deduce other useful properties like the health of your hardware (leaks), or the performance of the new parts (how much more air does that new intake really flow?).

 

Now what I gathered from that well written explanation was: don't waste your time trying to get it perfect, get it within five or ten percent and then let the adaptive system do its job. Which has been my line of thinking all along, correct me if I'm wrong.

 

The PCM uses the NBO2 sensors to perform closed loop correction of the AFR (keeping it at 14.7), as you noted.

The NBO2 is able to accurately/precisely operate only within narrow AFR range, something like 14.3-15.1 (or similar)... outside of this range the NBO2 has too much variance and can not be relied upon to produce the same result each time and from NBO2 to NBO2.

When you wish to accelerate, you open the throttle wide, and the PCM commands a richer AFR, something like 12.5-12.9... this is outside the operating range of the NBO2, so the PCM can't use the NBO2 to perform CL trimming of the AFR, the PCM is said to be in OL.

Now, regardless of running in OL or CL, the PCM calculates the cylinder airmass from either of the MAF or VE; it then determines the AFR to command (either 14.7 for CL or from table lookup for OL); it then computes the fuelmass required to produce this AFR;

then:
if in CL the PCM adds the previous trim to the calculated fuelmass (which keeps the AFR at 14.7);
if in OL, the PCM adds the previous trim only if it was positive, otherwise if the trim was negative the PCM ignores it;

this means that at WOT in OL, if the previous CL trim was positive, the PCM keeps adding fuel, or if the previous CL trim was negative, the PCM stops adding fuel;
so right there is one reason why you want the MAF/VE tables correct, so that the trims can be close to zero, so the PCM doesn't significantly modify fuelmass during OL;

in OL, the PCM relies upon the MAF and/or VE tables being correct, and if they are correct, the AFR produced will equal what the PCM is commanding (from the lookup table);

this makes dialing in the best AFR at the dragstrip an easy/repeatable task... you set a particular AFR, and because MAF/VE are correct, that AFR is produced; you note the ET/TS and adjust the AFR for the next run in an attempt to improve the ET/TS;

another reason is during airflow/airmass transitions (you significantly moved the throttle), if the MAF/VE and injector tables are correct, the produced AFR will be exactly track the commanded AFR, so the engine performs as commanded (i.e. avoids lean/rich stumbles/bogs);

another reason is that spark timing lookup is performed based on calculated airmass; if the airmass is wrong, the PCM looks up the wrong spark timing (engine will either detonate or not make optimal power);

another reason is that auto trans line pressure is commanded based on calculated torque which is computed from cylinder airmass (among other things); if the airmass is too low then the trans may slip.

 

Thanks for the explanantions, I should mention that I am running a CLSD tune with LTFTs turned off, so it's basically the VE table working the fueling.