2020+ Neural Network VE tuning on Silverado 1500 6.2

Hi,
I have just started tuning the neural network tuning system on the 2020 silverado 6.2. I have a supercharger with water to air intercooling fitted to otherwise stock internals.
Issues so far:
The 2020 silverado has a LIN BUS enabled MAF sensor. Barometric pressure, Intake air temp and probably a few other things are sent via LIN Bus to the ECM, the old "swap the intake temp sensor to the LT4 style MAP+Temp sensor" doesn't work. I have had some success modifying the manifold air temp model to bias more towards coolant but obviously depending on the level of heat soak in the intercooling system this doesn't read actual intake temp which confuses the model a bit. Anyone found a better solution for this? Is it possible to wire in the factory sensors to the E90 to enable SCIAP and temperature?

Having said that the stock cal holds together really well. I did have to re-tune the MAF curve at the top end (using stock airbox and MAF) as we are near the end of the range of the standard sensor and obviously the spark and fueling to get knock and EGT's under control but that was straightforward.
As far as VE tuning goes I have stock parameters which are "Volumetric Efficiency Aifrlow" and "Mass Airflow (Sensor)" I can just compare these two, because I know the MAF is good (comparing steady state LTFT+STFT and Widebands) so if I now tune the VE airflow to match the MAF airflow I should be right? Is this the method others are using? I don't see that I need to fault out the MAF etc and tune open loop as the older Gen4 system if I have these variables.

Also the neural network is pretty tricky. I just want to target the high load mid rpm VE range which is overpredicting, most of the VE airflow looks good, but it re-calculates the whole curve coefficients so I am always changing the whole map. Each time I spin the online calculator the numbers change a bit even with the same inputs so it's tricky to know if I am fixing my problem or causing more problems in other areas. The load axis is also just marked "Load" with 0-1. I changed this to 1.5 because I am roughly putting in 50% more air, but does anyone know what parameter is acutally used to calculate "load" is it cylinder air? Likewise with intake cam position, it appears to be referenced to something other than "degrees of advance."

It would be great if HP tuners had some method of predicting airflow from these tables or some sort of compare function.

Anyway if anyone else is having success using the neural network tuner on GM engines I would love to hear your thoughts.

The load axis is in grams/cylinder. Also use (2321) for the load and (2172) for cam angle...

Do this mean that GM made a neural network for the 2020+ ecms like dodge now,?

Is this "neural network" thing just a fancy marketing term for some kind of automated VE tuning or something?

Quote:

---

Originally Posted by cmitchell17

Is this "neural network" thing just a fancy marketing term for some kind of automated VE tuning or something?

---

no, its a different logic altogether. dodge has been using it for years.

I just don't get how it applies to engine controllers? I guess engine controllers could do better if they "learned", but I would think that this would require a lot more memory than the 2MB or whatever the flash has.

https://www.hptuners.com/products/neuralnetworktrainer/

Training a neural network is a resource intensive task. Running a neural network, not so much. The ECU's can run the neural network, but they can't adapt to changes made to the engine. That's why tuners need to retrain the neural network after engine modifications are made.

I see there is an option for the T87A? I just think theres no way the T87A has this neural network thing given how crappy the T87 and 8L90 shifts haha

The neural network is used to estimate Volumetric Efficiency in some Dodge and GM PCMs. As far as I know, their TCMs don't use neural networks.

Victor....where is HPT referencing the camshaft angle values from in the trainer? They seem like arbitrary numbers and are not like prior camshaft angles which started at 0 degrees and retarded the camshaft from there...

Quote:

---

Originally Posted by Redline MS

Victor....where is HPT referencing the camshaft angle values from in the trainer? They seem like arbitrary numbers and are not like prior camshaft angles which started at 0 degrees and retarded the camshaft from there...

---

I asked one of our engineers who specialize in GM tuning about this. Will post an update once I get a reply from him.

Quote:

---

Originally Posted by Redline MS

Victor....where is HPT referencing the camshaft angle values from in the trainer? They seem like arbitrary numbers and are not like prior camshaft angles which started at 0 degrees and retarded the camshaft from there...

---

We added 2 new parameters with the latest VCM Suite Beta(4.9.389). "2077 - Intake Cam Angle from Open Position" and "2078 - Exhaust Cam Angle from Open Position". PID 2077 should have the same value as the "Intake Cam Angle" input to the neural network.

sometimes i feel like GM's cars are being designed by ex-NASA scientist

I've seen a few places where it was referenced that the load axis is PID 2321 Cyl Airmass... but when I pulled the table, I'm showing Pressure Ratio in the row axis. RPM is still the column axis. I was planning on using PressureRatio (Manifold Pressure / Baro), which is already a predefined math... but figured I'd check now to avoid chasing my tail later.

Has anyone had success with NN editing? I am making just a global 15 and it does nothing anywhere to fuel trims

Just did a 2020 cam truck for a dealership 2 weeks ago.. spent about 5 hours on it with NN updating (about 8 updates to mail it) fuel trims fairly tight and been doing good.. still may be missing some pieces to the puzzle but they can be tuned fairly good

Man it?s not liking any of it. He put a cam in it, and it?s running about 15 rich. Tried globally pulling 15 and adding just to see if it made a difference, nothing. I even messed with injector flow rate and nothing. Only hacking the maf made a difference, and I really didn?t want to do that.

I posted in your other thread… don’t hack a tune, leave IFR alone. Tuning a MAF system is needed on almost any modded vehicle so that’s acceptable