rwyatt365

After finding a solution to my backfiring situation - which turned out to be EAE-related - I decided to try to create an (cue angels singing) ultimate EAE setting for my car. I know, this is a foolhardy quest, but this is the Final Frontier for me (until I try to accomplish sequential spark, or sequential fuel, or making a full-return fuel system, or...).

Anyway, here's what I'm trying to conquer, an "annoying" rich/lean spike between shifts;


I've attached my tune, but here are the AtW and SfW curves, as well as the RPM and CLT adjustment curves that I'm working with. I've arrived at these over the course of 6 months or so of incremental changes and playing with any number of different shapes for all of the above.


Any and all critiques/inputs are welcome...

DNMakinson

iTrader: (3)

AtW and SfW vs Coolant are backwards. Think that a cold wall will condense and hold more fuel than hot walls.

For smoothing our after start effects, I depress AtW at cranking, rather than SfW after start. The major reason I have adjustments at high RPM was to try to stop the wagging of AFR at full throttle lift. I think the key is to match when all of the puddle really is evaporated. I think that was a fools journey, as the nuances of IM temps is just too trick to get exact. Otherwise, from about 4K up, I would not have adjustments. In other words, to actually predict and control the "Wall Fuel" parameter to match reality. The relative WF is not easy, the actual, I think is much harder. @Ted75zcar says he has an off-line algorithm to set EAE, but I don't know all that he has done with it.

I think your RPM at which both should be at 100% should be around 3000, and that would be where you would define you base AtW and SfW curves (vs kPa).

Here is my discussion from sometime back

Here are my charts: The big bump in RPM at 1500 - 1800 was to try to correct pull away from stop leaning out when car has not been running long. It does not help a lot, as the effect is short lived. What helped more was to add WUE and stop EGO until about CLT > 120*F. (Thanks Curly) It tends to still lean a little, but remains in a good drive-ability range.





DNM

Ted75zcar

iTrader: (2)

While conceptually I agree that the CLT curves are backwards, the tuning algorithm I developed insists on having the curves come out inverted from what you would intuitively anticipate.

Keep in mind that as added goes down, WF goes down, so sucked which is a function of WF will go down too. Since we are dealing with an integral here, the 2D curves don't provide a really good visualization. So with really small wall fuel, we get almost no sucked contribution, and the full added goes to the wall.

Sorry for the crap pic, my engine cpu won't network at the office.

DNMakinson

iTrader: (3)

Ted, worth playing with, especially as you and Roger got similar results.

It has been a long time since I played with EAE, but my recollection is that my settings worked well. One problem is that the car spends little time at low temperature. I will flip mine and see what performance looks like, then flip them back. I will also record wall fuel.

EDIT: Roger, I did nothing toward your question of the Rich / Lean, as I'm discussing your CLT tweaks. I think that your basic kPa curves are not yet aggressive enough. I may be over-compensated, but I tend to go Lean / Rich on a shift.

Ted75zcar

iTrader: (2)

With 6 curves, I suspect that one could come to a configuration that works well in many different ways.

DNMakinson

iTrader: (3)

True. But as a matter of course, the tweak curves should cross 100% at the primary tuning points. If that is 185F and 3000 RPM, then the RPM and temp corrections should be 100% at that temp and Rev. Otherwise, one chases one's tail.

Ted75zcar

iTrader: (2)

I started with a single nexus point in the way back of 2500-40-185, all curves used to servo about this point. I later enhanced the algorithm to allow each curve to servo about its own center or 100% point.

rwyatt365

So, I'm hearing some takaways;

1 - Invert the CLT corrections for both
2 - Cross 100% on the correction curves at the "prime operating points" (185F and 3K RPM)
3 - Get a little more aggressive with the kPa curves

I'll try these, in sequence, one at a time to reduce "crosstalk". I have some questions/comments about this...

For #2, If I shift either of the CLT curves wholesale, by some delta, shouldn't I adjust the corresponding kPa curve by the inverse (i.e. if I pull all of the points in the CLT curve down by n% so that it's at 100% for 185F, shouldn't I multiply the corresponding kPa curve by 1.n)?

For #3, I daren't go above a SfW of 10% in the upper kPa values or I'll get back into the "backfire" condition from my other thread. I think I can get more aggressive with AtW, and I'll give that a try.

So...tomorrow morning will be to try #1.

rwyatt365

Ok, first iteration this morning. I inverted the CLT compensation curves. Impressions? The only thing that I "felt" was that the behavior coming back to idle after being on-throttle was better. Where before there was a noticeable dip in RPMs before coming to a stable idle, today that dip was less pronounced. What does the data say? After comparing today's log with yesterday's, there seemed to be no real difference between the rich/lean spikes between shifts (that I'm trying to address). But I can see that the wallfuel is significantly higher everywhere. That's telling me that I'm working with a bigger "puddle" in the IM than before - we'll see if I can work that to my advantage with regards to the lean part of the inter-shift spike.

Now, here's the data;
"Inverted" AtW & SfW CLT curves


Today's log compared to yesterday's. Wallfuel is red in the lower plot.


This afternoon I'll shift the SfW CLT curve down 20 "points" so that it's at 100% at 185F - you can see that it's at 120% between 180 & 190. I'll also shift the SfW MAP curve up by 20% (multiply by 1.2) to compensate for this change. On second thought, maybe I'll do this in two "chunks"; shift the CLT curve down and log that, then shift the MAP curve up and log that ("scientific method" you know, change one variable at a time).

DNMakinson

iTrader: (3)

I did try the inversions as well. Will probably be able to post tonight. But, yes, the shift thing is at normal operating temp, so the temp corrections should have little to no effect.

Personally, I did give up on trying to get those extreme throttle lift lean-outs to go away. In the end, you are looking at AFR during a time where you are in extreme vacuum in the IM, and are asking the engine to make Zero Power. It has no effect on drive-ability, engine heath, etc. If you did not have the gauges / plots, you would never know they existed. I do believe this is one of those walk-away things.

That is, as long as the lean spike is actually occurring when the throttle is closed, which I fully believe is the case.

What concerns me more is the strange inversion on throttle depresses before and after the cyan line in your above chart.

Really, EAE does not need to be perfect to give the desired result of keeping AFR's on target for most transient situations.

DNM

rwyatt365

I know...but those spikes jar my aesthetic sensibilities (this coming from a guy with a purple Miata!).

Anyway, here's a compressed view from the same log where there were some lower-tpsdot throttle manipulations and you can see that EAE is doing a remarkably good job of keeping the AFRs relatively stable;


Every throttle lift is accompanied by some degree of rich-trending AFR - depending on how rapid the lift is (i.e. negative tpsdot). I'll be doing the SfW CLT shift shortly (almost lunchtime and before the rain comes).

Ain't science FUN?!!

rwyatt365

So, I chickened out and only reduced the SfW CLT by 15 and not 20 "points". The result was that the drivability suffered from the standpoint that the that the off-throttle idle droop has returned. Beyond that, the AFR at idle is richer than before. The data is showing that the rich/lean dip at the shift transitions is "deeper". What I mean by that is that the rich dip is richer, and the lean spike is less lean an seemingly "sharper" in profile. Because of traffic, I couldn't get a representative number of examples, but here's one that shows what I'm talking about.


Tomorrow, I'm going to reduce the whole SfW MAP curve by 15% and see what happens.

DNMakinson

iTrader: (3)

OK, I'm trying the opposite. I multiplied my entire SfW by 1.1, and got less of the activity you are referring to, up to about 3500 RPM throttle lift. Theory is that the Wall Fuel (calculated) is held smaller, and thus better matches reality. Then upon lift, there if also more SfW which stops the rich jump, and then the calculated WF gets depleted, just like the real one, and thus the lean is properly accounted for and therefore eliminated.

The trick with EAE is that you are not telling the engine what to do, rather, you are trying to "know" what the engine is doing so MS can counteract the bad behavior. In other words, you don't cause the SfW, you just try to code for what is actually happening as kPa goes down. If you code it correctly, then the ECU correctly knows how much fuel to pull from each pulse to account for the extra fuel coming off the walls and going into the cylinders.

Again, getting the relative amount of Wall Fuel is tough, but getting the exact amount, especially predicting when WF = 0, or is at a true minimum and not changing.... that is REALLY tough. I think that is what you are after.

In the AM, I am going to turn up SfW RPM multiplier from 4K and up and see what that does. Dang, you have me working on something I was not going to fool with.

EDIT: Of course, on a throttle lift, the RPM is changing at the same time the kPa is changing. That is also an additional complication to this problem that does not occur with a change in TPS / MAP when the engine is tied to the transmission.

DNM

rwyatt365

There were some typos in my previous post (which I edited). Anyway, here are the "differences" between today's run, and my baseline in the first post;

SfW CLT is shifted down by 15 points across the board, and SfW MAP is multiplied by 1.15 (15% increase) except for the last two points, which I'm intentionallykeeping below 10% so that I don't re-introduce the high-boost backfire condition.

Here's what that looks like in a datalog;


The difference is subtle, but it seems like it's trending to what David was speaking to - that WF is properly accounting for the throttle lift (to a degree) and not going as rich AND that it's being sucked off the walls enough to (almost) deal with the lean spike. This is encouraging, and heading in the right direction, I think. The question is how far can I push this to get to a flatline AFR?

One of the problems, as David mentioned earlier, is that all of this is RPM-sensitive and you can see that I didn't hit the same RPMs between the different logs. That makes comparisons tricky because it's hard to differentiate between what's a result of the MAP curves, and the RPM curves. So, what you're seeing above may just be because there's difference in the RPMs.

Tomorrow's quest...can I bump the SfW curve up more and keep the trend towards a flatline AFR going? Tune in, "News at 11"!

Ted75zcar

iTrader: (2)

My testing indicates that there is a much higher dependence on RPM than map.

I don't think you ever get rid of the rich lift condition. You can't remove fuel that has already been delivered.

rwyatt365

I spent a few hours months ago when I first started this quest to find some "magic calculation" that I could use as an "index", or "factor" of some sort that I could use to help guide me to a conclusion on how to adjust all of these curves. You know; MAP * RPM / WF, or a "magic" scatter plot that would lead the way. But none of that made any logical sense, and it didn't point me in any rational direction so I gave up.

If anyone has any thoughts, please chime in. Everyone needs an "EAE Silver Bullet"!

Ted75zcar

iTrader: (2)

I spent many hours, well over a hundred for sure, developing an EAE tuner in MATLAB. It incorporates 4 separate machine learning models, any many hundreds if not thousands of lines of code. From this experience, I think I can say with some degree of confidence that there is no easy equation for this stuff.

DNMakinson

iTrader: (3)

I will say again, however, that getting close, and improving drive-ability, and fixing what matters is an attainable goal. The lean spike after lift is very tricky, and is of no import.

In fact, I was running a very short time delay on fuel cut and if I took a long time to shift, I would trigger it. Then go full lean. Did not feel any different from when it did not occur.

I agree that RPM is playing in and is quite tricky.

I have tried boosting SfW at higher RPM and think there was some improvement. Also, noted that my pulse was getting into the 1.3 range, which is non-linear on FF 640's. So, I have turned on the NL small pulse correction. Truly, when we are running something like 17 kPa, it really is only small amounts of fuel to make a big difference in AFR.

EDIT: Roger, how lean are your lean spikes? I cannot tell.

DNM

Ted75zcar

iTrader: (2)

Another key is to tune the VE tables in the decel rows. This is tricky because of a bunch of stuff, including the small PW that DNM mentioned. It really helps to be looking at SD logs here, I recommend a 5ms sample rate.

Edit: and it looks like you have heavy filtering on MAP, this is hurting you if you do.

rwyatt365

My goal is to maintain (or improve) the current, low-boost drivability while "making the logs look pretty" - typical ---- retentive stuff. I'm pretty sure that the lean spike is not a mountain I'm willing to die on. When I look at the numbers, and not the shape, the spikes are not really that bad. Some, of course, where fuel cut turns on go full-lean. But others tend to be between 14.7 and 15.2, which - by the numbers - isn't that bad for a transient event under no load.

My PW will go down to 1.2-ish which is low...maybe too low to accurately control as you mention. I'm working with a PNP2, so I don't have all dem-dare fancy schmancy corrections to adjust for that.
I've got a 10 kPa row in my VE table, but I've virtually ignored it recently. I'm ashamed to say that it's original purpose was to induce some burble into the exhaust on decels, but I quit that adventure a while ago. What would I be looking to do here? My guess is to add some fuel to bump up the PW before fuel-cut comes on.

And, if I have filtering on MAP, then I don't know how it got there, or how to do anything about it - soliciting instruction. Remember, this is for a PNP2, so I have LOTS of limitations.

Next Steps;
So I guess my next target is those pesky "rich dips". My operating theory with those is that WF is high on the run-up to the closed throttle event so that, when the throttle is snapped shut, that fuel is sucked off and everything goes rich. Since the consensus is that RPM has a more dominant effect than MAP, my plan is to overlay shift events from the same log where the MAP is similar but the RPM is different and compare what the AFR is doing. That should show me what needs tweaking at that spot in the RPM tables...and I'm thinking; first the AtW RPM table, then the SfW RPM table (since SfW is a percentage of AtW). Does that make sense?