Update, for building an error graph, HPTuners added a new parameter that matches the neural network. From Hp Tuners - 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.
worth mentioning that on a GM file, the "load" isn't measured in absolute load (%) but in cylinder airmass (g/cyl). Also, when you are building your histogram for different cam angles you would want to set the filter, for each cam position, to be something closer to halfway between the two angles so for example for cam angle 117 you would set the filter for (111.5
Trying to train the NN now with a 6.4 Hemi ram but the NN table that's 'Pratio x rpm' for editing, also has 5 intake cam and 5 exhaust cam angles available in the drop down options. How you go about the process with 5 intake cam angles and 5 exhaust cam angles? Did you make 25 different 'PRratio x Rpm' tables? I understand there is only 1 cam in these and it's the exhaust phaser doing the work, but I'm lost after that. Thanks
Man I would love to jump in to NN on my ram...im just kinda bent out of shape about the 50 bucks a month thing after i spent 600 bucks on the mvpi2 and credits. At least with dodge you can elect to use either NN or ve
THANK YOU SO MUCH FOR THIS VIDEO!!!!!!!! QUESTION: Are the Mopar Nnets for engines with VVT? Would 1996 Viper GTS (Gen 2) NOT have Nnets ? OBSERVATION: I've spent over 15 years working with Nnets and I couldn't understand what this Nnet is doing mainly because I could not read the various parameters and headings in your screen shot. It sort of looks like there are max/min specs for input and output layers? Is there any specification of # of hidden layers or # hidden nodes? I assume you cannot add parameters to the input or output layers? But now I get to the real question: Where is Training Data coming from? To train a net you need a Training Set which is a set of various engine states where you _know the answer for each state._ Then when you have a new input that has never been seen before, the Nnet produces a good guess for the outputs. If you mod an engine, the relationship between inputs and outputs (MAP in particular) is changed requiring a different training set - remember when you train the net you need to tell it the correct answer or output. It seems like there is a ton of stuff going on "behind the scenes."
Gen 2 does not have it. They are training the new network based on the data that you provided from generating a new volumetric efficiency table that takes into account the AFR errors that you log, so effectively you are saying based on the known data this is the correct outcome and then they run it through their backend and generate new weights. There isn't a way to change the inputs as far as I know you're just correcting for errors that are outside the original training set.
@@GoatRopeGarage Maybe you could help me: I'm trying to find someone who codes JTEC to pull some timing when A/C is turned on. I code but not automotive. It seems it should be a few lines of code. I also wonder if all the unused pins are available when writing code?
Thanks for this. There is always missing/gaps in data in the table when you build and log the error to the histogram. What do I do with this missing data? Do I enter zeros? Leave it blank? If I leave it blank does the backend of the HP tuner system take care of this for me?
So if I am going to be installing a supercharger, would I increase the NN "VE table" by the boost percentage above atmospheric, or just keep adding 10% at a time until it runs stoich?
I did it on my get 3 using towhual/performance mode. Uses a single ground pulse in and out to turn off and on, gotta change activation settings but super easy, should add its basically only transmission tuned but it makes all the difference
I don't know but it seem that hp tuners could integrate the trainer in VCM editor instead you must pay to be sent your VE table file to open and dial your VE table
What would cause the B2 Upstream O2 sensor to constantly on the graph going up and down while the B1 is pretty stable? Already replaced the B2 O2 sensor?????
Bouncing between 0-1000mV is correct operation for narrowband O2 sensors reading around stoich. I think maybe you should replace the B1 sensor if it’s stable?
I had my stroked 392 tuned in December and the tuner turned off the Neural Network...said you have to with a cam....I have a limiter installed......with the network turned off this doesn't let the 7 degree usable on the cam right?????
@@mysixfo8805, a lot of the aftermarket cams didn't used to support VVT. For all the years up to this addition, you weren't able to tune with neural network turned on, there wasn't software to do it, so you had to turn it off and go to standard VE tables. Disabling the neural network eliminates the VVT.
If that won’t work, another old hydraulic trick is to cut a short piece of angle iron at the length you need and hose clamp that to the cylinder. My welding chair is still rigged that way to this day.
Backed up by experience. Been tuning PID loops for 20 years, that's all standard closed loop is. I know there limitations of closed loop and the possibilities of neutral networks.
@@GoatRopeGarage I understand what neural networks can do I just don't understand what they're doing in HP tuners. Maybe it's because I've only done ford stuff