Pedal Percent Power Request Modes

[Stoopalini]

Not sure what you mean. How do you figure I?m telling the trainer the LSA is 114 degrees?

[LilSick]

I am out and about and on my phone, but..

The lobe separation angle is ground into the camshaft and is confirmed when you do the math to determine how many degrees before top dead center the centerline of the exhaust lobe is at, added to how far past TDC you determined the centerline of the intake lobe to be, divided by two.

118+111 = 229

if you divide that by two you snd up with 114.5, the actual lobe separation

[Stoopalini]

It’s 228.5 for the 5.7 OEM cam. Even the OEM cam settings confirm this (ie: compare the stock exhaust and intake base or WOT VVT tables and you’ll see that each intersection totals 228.5)

But even so, not sure how I am telling the trainer what the LSA is. The trainer is actually informing me of it due to having fixed values … meaning, you cannot enter whatever you want for cam positions.

Also, if yiu look at the min and max settings, yiu can also see its 228.5.

Curious, what makes you think it’s different?

[LilSick]

It?s 228.5 for the 5.7 OEM cam. Even the OEM cam settings confirm this

not sure how I am telling the trainer what the LSA is. The trainer is actually informing me of it due to having fixed values ? meaning, you cannot enter whatever you want for cam positions.

ha. what it says in the tune does not confirm anything. hard to believe 5.7 is over 6 deg tigher than a 6.4

i am not even certain 6.4 is 120.5

gonna find out tho:

[Stoopalini]

Those drop downs in the neural network trainer have fixed values, 5 in each drop down for a total of 25 possible tables. But since the 5.7 has a single cam, only 5 of the tables are actually valid.

So when selecting 117.25 for the exhaust cam position, only one of the intake drop down selections can be valid.

The options for intake are

• 100.5
• 111.25
• 122
• 132.75
• 143.5

Options for exhaust positions are:

• 85
• 95.75
• 106.5
• 117.25
• 128

The only LSA which makes sense based on those numbers is 114.25, which seems to be confirmed in several other places throughout the tune. I guess if it?s wrong, seeing as the factory calibrations are using it, it?s probably a safe bet to continue using it.

[LilSick]

yeah i dunno

someone might put a degree wheel and a dial indicator on a 5.7

see where it is at full advance, then take the spring outta the phaser and see how far it will retard?

[Stoopalini]

Yeah, I understand you're frustrated no one has degreed a factory 5.7 cam yet .... but in the meantime, I think working off the assumption the NN trainer and the factory calibrations are accurate using an LSA of 114.25 (228.5) is a safe bet.

In terms of dialing in the NN virtual VE tables above 1.0 PRatio, the LSA really shouldn't have a huge impact anyway. My current LTFT+STFT trims are looking great, it's only the EQ error when in PE mode, and above 1.0 PRatio in the upper RPMs, where things lean out.

I think it's related to the NN scaling I did, and the trainer didn't populate the VE values in the new cells very well.

Here's the EQ error with cam between 112 and 122:
EQ_Error_Cam-117.JPG


Here's what the virtual VE table looks like for 117.5 to produce that EQ error chart:
ANN_Cam-117.JPG


And here's the SD VE table the 1st tuner I used created when he had ANN disabled:
SD_VE.JPG


So it looks like the NN just didn't scale out well to the additional Pressure Ratio rows I added, as well as the upper RPM for the existing rows. This makes sense to me.

I think I need to adjust those areas in the virtual VE tables, and then smooth it out and finalize the training to produce a new set of weights and biases. I just haven't done interpolation and smoothing on VE tables yet, so need to figure out the best way to do this.

[LilSick]

"So it looks like the NN just didn't scale out well to the additional Pressure Ratio rows I added, as well as the upper RPM for the existing rows. This makes sense to me."

once you are accelerating i do not believe it makes any difference which position you have the switch in [2088]

that os does not incorporate wide band sensors

it would be different if it did

[Stoopalini]

2088 is the switch to enable or disable ANN ... Are you saying ANN does not adjust fueling when PE is active, on vehicle which do not come with a factory wideband?

If that's the case, then where is the control mechanism for fueling during WOT if it's not the ANN?

I thought ANN would adjust fueling during WOT, but do it blindly without a feedback mechanism. Therefore, the virtual VE table tuning is critical because the PCM cannot adjust for variances in fueling when something other than lambda is being targeted.

So ... closed loop, lambda target, upstream narrow band O2 feedback loop drives fueling adjustments through STFT and LTFT. But when open loop kicks in, and the PE table is being used to target something other than lambda, then the narrow band O2 signals can no longer be used to make adjustments ... So the PCM blindly uses the ANN to control fueling with an attempt to hit the non-lambda target. Although without a factory wideband, there's no way for the PCM to know if it's achieving the target or not.

In my case, it's obviously not. So the ANN virtual VE tables are just a way for a user to try and understand the ANN's weights and biases and make adjustments to them.

If my understanding is incorrect, then where do I make the adjustment to add more fuel in the upper RPM and PRatio ranges?

[LilSick]

PE

when the JPEC2A controller goes into PE it is no longer using the neural network or lambda (or trims imo)

it is using a calculated fuel air ratio

your tune is selecting the pressure ratio table [34243] because of how you have [34229] set

PE table + (rpm vs pressure ratio adder)

your stoich FA is .0709

with the adder in PE it increases to .0821-.0918 depending on where you are on the table

you can divide 1 by those to find the air fuel ratio but its all just numbers lol

the calculated grams of fuel the controller is trying to supply the engine with keeps increasing as rpm increases

just as it should according to the table [34243]

there is no EQ error cuz it is not referencing lambda, it is in PE, which is a calculation that spits out something that is converted to a pulse width

you have a wide band so you can watch how close the mixture is to lambda all thru the rpm curve when making a pull tho

i would like to see you do a compare, using a phone hanging from your sunroof to watch a good controller (like a ballenger) that was not wired into the CAN or going thru scanner

point it at the ballenger and the tach

and dont do a launch... lock it in 3rd or 4th and run it from 2250 to redline

and hook up the pressure gauge

lets see how good your fuel pump is keeping up

increase the numbers in 34243 and it will increase the fuel mass (pulse width) when in PE

if your mixture is not fatter after increasing those numbers, then there must be a supply issue

[Stoopalini]

PE

when the JPEC2A controller goes into PE it is no longer using the neural network or lambda (or trims imo)

it is using a calculated fuel air ratio

your tune is selecting the pressure ratio table [34243] because of how you have [34229] set

PE table + (rpm vs pressure ratio adder)

your stoich FA is .0709

with the adder in PE it increases to .0821-.0918 depending on where you are on the table

you can divide 1 by those to find the air fuel ratio but its all just numbers lol

the calculated grams of fuel the controller is trying to supply the engine with keeps increasing as rpm increases

just as it should according to the table [34243]

there is no EQ error cuz it is not referencing lambda, it is in PE, which is a calculation that spits out something that is converted to a pulse width

you have a wide band so you can watch how close the mixture is to lambda all thru the rpm curve when making a pull tho

i would like to see you do a compare, using a phone hanging from your sunroof to watch a good controller (like a ballenger) that was not wired into the CAN or going thru scanner

point it at the ballenger and the tach

and dont do a launch... lock it in 3rd or 4th and run it from 2250 to redline

and hook up the pressure gauge

lets see how good your fuel pump is keeping up

increase the numbers in 34243 and it will increase the fuel mass (pulse width) when in PE

if your mixture is not fatter after increasing those numbers, then there must be a supply issue

What you've outlined here is how I initially understood things to operate, which is why I previously mentioned increasing the PE table to address the fueling getting leaner than desired in the upper RPM/PRatio. But HaasEXP said the PE Tables were used for the target, not actual ... and thus the ANN needed to be tweaked. Seeing how the ANN virtual VE tables drop below 100 above 5k RPM, and down into the 80's at 5500 and above ... it seemed a reasonable suggestion.

I am logging FA Enrichment, and can see 0.0209 is being commanded as expected from table 34243

.0709 + .0209 = .0918
1/.0918 = 10.89 AFR or 0.77 Lambda

I'm using 10.89 / 14.1 to calc lambda because as you mentioned, I have stoich set to .0709 (ie: 14.1 to account for E10), and this aligns to what VCM is reporting as well

So commanded is 0.77 lambda, which includes the PE .0209 adder, but I'm only hitting 0.86 lambda at 5600 RPM. If I increase 34243 to add more fuel, I imagine it would work, but would also result in commanded being reported even fatter than 10.89/0.77.

And when I look at the virtual VE tables and compare them to the SD VE table a previous tuner created for my vehicle, the delta seems to line up.

So I think it's worth trying the trainer 1st, before going above .0209 PE; although I'm travelling this week for the holidays so probably won't get a chance to try it until the weekend or early next week ... Unless I get some time later today

[HaasExp]

I thought ANN would adjust fueling during WOT, but do it blindly without a feedback mechanism. Therefore, the virtual VE table tuning is critical because the PCM cannot adjust for variances in fueling when something other than lambda is being targeted.

So ... closed loop, lambda target, upstream narrow band O2 feedback loop drives fueling adjustments through STFT and LTFT. But when open loop kicks in, and the PE table is being used to target something other than lambda, then the narrow band O2 signals can no longer be used to make adjustments ... So the PCM blindly uses the ANN to control fueling with an attempt to hit the non-lambda target. Although without a factory wideband, there's no way for the PCM to know if it's achieving the target or not.

Your understanding is correct. the PCM enters "open loop" which means it still references the ANN VE tables (because it is still VE based) but there is no active feedback from the fuel trims to trim the IPW and thus correct the lambda during open loop, unlike in closed loop operation. The actual EQ error in this case is the actual lambda (measured by your wideband)/target lambda (1/[stoich FA + PE adder]). This obviously assumes that no LTFT+STFTs are held into WOT (this does happen on some GPEC2/A PCMs).

To get good data to tune the ANN at WOT, are you recording multiple logs with fixed WOT cam timing that relates to the virtual VE table you are actually trying to tune? Eg, if you are tuning the 117.5deg virtual VE table, are you fixing the cam angles to 117.5deg? If not, then you will get a mix of airflows that will skew the actual error and actual VE for that particular cam angle.

[LilSick]

Your understanding is correct. The actual EQ error in this case is the actual lambda (measured by your wideband)/target lambda (1/[stoich FA + PE adder])

why is what you see in these pics allegedly happening

according to the log it happens during the convertor stall

happens over less than a quarter of a second

happens over a 200 rpm increase

the error is swinging back and forth from 50% lean to 20% fat

i know scanner shows it happen, but is it actually happening?

[Stoopalini]

Watch the RPM through that range of the log ... the tires were spinning there, very slightly, but still happened. You can watch the RPM drop for a split second.

I had the vehicle in Track mode, with stability control completely disabled.

[LilSick]

i call bs

again lol

you do not log wheel speed, but look at trans output shaft rpm...

its increase is linear and consistent

i added it to chart vs time and made its high reading 1100

zero tire spin - just the torque converter slipping while the vehicle caught up to the stall speed

you sure you can trust your o2 data?

i have a hard time seeing the mixture fluctuate like that

especially by that amount!

the first lean spike, the one at the hit... that one might be real

[Stoopalini]

I normally don't see that anomaly in my data, but I typically do not do 0-60 tests in track mode with stability control completely disabled either. I felt the tires slip/grip when I punched it. Not tire squealing wheel spin, but it definitely was fighting for traction after the initial second or so of lag had passed.

[Stoopalini]

So I made some ANN tweaks to the 117 and 106 exhaust cam virtual VE tables, specifically in the upper RPM range and especially above 1.0 PRatio, and did a re-train with only the 5 valid tables selected.

Due to where I am right now (travelling for Thanksgiving), I can't really do a 0-60 pull, but did run out to pick up dinner and logged the drive. There's a few times in the log where PE was engaged, just briefly ... and the EQ tracking seems to be improved. I did notice it carried positive STFT into one of the PE areas at one point, so need to double check this when I can do another log with some WOT opportunity. Maybe I need to add some more to the areas preceding this (ie: lower RPM and lower PRatio areas_... But it does seem that using the NN trainer did impact the WOT fueling. I have PE WOT threshold set to 2.6v in field 44442.

New Log: TH_11262024_Base_ANN-Adjustments_After-training-2.hpl

[Stoopalini]

So I'm finally back home, and am working on 5 test tune files to capture EQ error at the specific cam positions for the ANN trainer.
I think I have them setup correctly ... I set the Exhaust and Intake VVT tables to be fixed, and also disabled DFCO and Component Protection.

I adjusted the following tables accordingly for each test file:

• Desired Angle - Normal (36554 & 36547)
• Desired Angle - Max TIP (36555 & 36548)
• WOT Desired Angle - Base (36559 & 36551)
• WOT Desired Angle - Max Airflow (36561 & 36560)

I'm not sure if I needed to do the Max Airflow tables though....


My plan is to load each tune, take a short test drive with various PT operation, and then perform a WOT run as well. I'm thinking the WOT should probably be a rolling run, maybe in 3rd gear from say 2500 - 5500 RPM.

Any risk to doing this with the VVT tables set to fixed numbers like this though?

Here are the values I used:

• Exhaust = 85.0 | Intake = 143.5
• Exhaust = 95.8 | Intake = 132.7
• Exhaust = 106.5 | Intake = 122.0
• Exhaust = 117.3 | Intake = 111.2
• Exhaust = 128 | Intake = 100.5

The numbers aren't exactly what the trainer uses (ie: 117.3 in my test file instead of 117.25 in the trainer) because VCM Editor won't let me enter two decimals of data into the VVT tables. It will let me extend the fields to more decimal places, but when I enter data, it keeps rounding it to only one decimal place. I figure for this exercise, 0.05 variance in two of the test shouldn't really matter though.

I also realize the intake numbers probably aren't needed, but I entered them anyway just in case there's any calculations being done in other areas based on the intake cam values. I've actually been doing this for all my VVT changes, just for completeness.

Here is the tune file I'm basing the test files off of: TH_11262024_Base_ANN-Adjustments_After-training.hpt

And here are the 5 test files:

• 85.hpt
• 95.hpt
• 106.hpt
• 117.hpt
• 128.hpt

Is running a test with 85 or 128 being commanded going to cause an issue, considering these values are not achievable? Is it even worth doing?

Or should I just run the tests with the 106 and 117 files, then only train with 2 tables checked?

[LilSick]

your blower is pushing compressed air into the cyl (while the intake valve is open)

that valve starts to open on the exhaust stroke...

it does not close til part way up the compression stroke

advancing the exhaust 14 degrees from straight up advances the intake by the same amount

which will close the intake valve 14 degrees sooner

that will increase the effective compression stroke

which raises cyl pressure.

if the cyl pressure is too high for the fuel your are running you will see detonation (even with the little timing that is present in your tune)

if the cyl pressure is too high for the hardware it lift the cyl head which will blow the gasket

if the cyl pressure is too high for the head gasket it will leak which will blow the gasket

[Stoopalini]

Yeah, thanks ... that's my point on why I was concerned about testing with the 128 degree file. Not only is 125 the maximum advance the 5.7 will do, but running WOT up to redline with it fully advanced would produce excessive cylinder pressures ... and I'm not even sure the data collected would be worth it, considering the data would be with the cam at 125, but the ANN virtual VE tables would be expecting the data for 128. If I took the data captured with the cam at 125 and entered it into the NN trainer's 128 degree virtual VE table .... I think it would then be off at 125 due to the trainer extrapolating what it thinks 125 should be. As it stands right now, LTFT+STFT and EQ Error look great when the cam is at 125.

I have similar reservations about running a test with the cam set to 85. T maximum retard of the 5.7 phaser is 97 degrees. So if I put 85 in all the tables, and tested ... the cam would only move to 97. So I imagine the data collected here wouldn't be worth inputting into the 85 table, due to the same thought process I described above .... because if I did take the data collected with te cam at 97 and input it into the NN trainer's 85 virtual VE table, then I think it would extrapolate incorrect data for the real 97 degree position ... and, the cam will never be retarded this far anyway.

I think the best course of action is to collect data with two tests. One having the cam locked at 106.5 and the second having it locked at 117.25. Then use this data to fine tune the NN through the trainer. Considering I am only swinging the cam between 112 and 116, I think this would work well. The cam does park at 125 at idle, but the trims and EQ error are very good during this scenario. Then any increase in PRatio or air charge causes the cam to follow the PT or WOT VVT tables, where I have it set as 112 min to 116 max.