Results 1 to 15 of 15

Thread: HPTuners Neural Network Training tool Dodge

  1. #1
    Potential Tuner
    Join Date
    May 2020
    Posts
    1

    HPTuners Neural Network Training tool Dodge

    I have recently installed roots supercharger(M122) on my 2014 Charger R/T and while prepping for the project I did quite bit of reading on tuning newer Dodge vehicles and how Neural Network can throw a ranch into the tune. Some people advised to disable NN (Neural Network) and do VE table tuning. Now that HPTuners came out with tool for NN training, I would like to document method of tuning while utilizing this tool. Has anyone given it a try?

    Based on what I understand, the tool trains NN based on new VE table. What would the process look like?

    1) Get base map, send it through initial steps of NN training tool to prep it for tuning.
    2) Disable NN? Keep it on?
    3) Datalog and correct VE tables
    4) Once VE table is solid, send it through NN training tool? reiterate as needed?

    Please share thoughts on this, thank you!

  2. #2
    Tuner in Training
    Join Date
    Nov 2013
    Posts
    23
    Anyone try it out yet? I haven't had a vehicle with an aggressive cam or the need to adjust, but I'd like to know how it's working out.

  3. #3
    Hi,
    You got two routes.
    First is to remain with ANN, and change variables (weight, gain)
    Second - oldschool to go with VE tabels. Please correct me if I'm wrong, but using ANN excludes VE tables. You cannot use both.
    During changes in my Chall, i went with VE tables.

  4. #4
    Pablito, you are correct, you can not use VE tables when using ANN. ANN makes decision off of air fuel ratio's commanded and conditions. VE, are based off of AFR, commanded and desired from what I understand (I'm aware that this is a very simplistic POV). ANN remembers conditions like humidity, temperature, barro, and takes all of these into account as it learns to create target fueling needed for the conditions. VE is way easier to deal with, as you can account for higher boost levels than ANN can at the moment. At least this is what I've found for the 11-14 GPEC controllers.

  5. #5
    Advanced Tuner f.creek-ranch's Avatar
    Join Date
    May 2018
    Location
    Germany
    Posts
    572
    Just for my understanding...
    If you got no widebands, the open loop uses VE tables for FA, how will the NN this handle without informations?

  6. #6
    Advanced Tuner
    Join Date
    Feb 2016
    Posts
    431
    ANN is there to make fuelling changes according to the cam position.
    The factory swings the cam alot for EGR and emissions...so much so that it kills the VE of the engine.
    ANN looks at the (often less than optimal) cam position and calculates how much more/less fuel is needed as the VE moves higher/lower.
    A fixed VE table works well if you have a stable VE engine (cam doesn't move or only a little) but if you have anything like
    factory cam tables trying to use VE will never be close as the engines VE is variable....So ANN is your best option then.

  7. #7
    Advanced Tuner f.creek-ranch's Avatar
    Join Date
    May 2018
    Location
    Germany
    Posts
    572
    thanks, that my opinion too.
    but what happens at open loop?
    The O2?s arent fast enough, so the software will play with fixed values.
    In this case the NN would only give the cam positions to correct the VE table or is the whole VE table produced by NN self???

  8. #8
    Advanced Tuner
    Join Date
    Feb 2016
    Posts
    431
    NN doesnt use a VE table in the traditional sense...
    It uses a series of inputs and weights to come up with a calculated fuel mass.
    During open loop (on a non wideband system) it has zero feedback so just outputs what it calculates.
    But don't think it is some kind of AI....its dumb and does what it is told to do.
    So even in CL the 02's are just adding/subtracting trims to the NN fueling.
    You watch... the trims stay stable... say +7%...only way to get it closer is to play with injector settings etc or mess with the ANN.
    Its not like it "learns" what the new fuel requirement is or the trims would go to 0....

  9. #9
    Advanced Tuner f.creek-ranch's Avatar
    Join Date
    May 2018
    Location
    Germany
    Posts
    572
    Its hard to play with neurons in my mind ...
    Im searching for something visible like the VE maps.
    I wonder how the OL works. I think the BIASES are the points of the VVT or the OL, because they have no input, only outputted function.
    How the hell can I mess with the OL if there is no map to correct.
    What will I do with the WB values that I log?
    If I do the 6.4 cam swap, how will the pcm will work correctly in OL?
    There are a lot of questions, I wonder why nobody discussed points like these...

  10. #10
    Advanced Tuner
    Join Date
    Feb 2016
    Posts
    431
    Just tweak the injector settings till you get the AFR's you desire....simple really and it works.

  11. #11
    Quote Originally Posted by Hemituna View Post
    NN doesnt use a VE table in the traditional sense...
    It uses a series of inputs and weights to come up with a calculated fuel mass.
    During open loop (on a non wideband system) it has zero feedback so just outputs what it calculates.
    But don't think it is some kind of AI....its dumb and does what it is told to do.
    So even in CL the 02's are just adding/subtracting trims to the NN fueling.
    You watch... the trims stay stable... say +7%...only way to get it closer is to play with injector settings etc or mess with the ANN.
    Its not like it "learns" what the new fuel requirement is or the trims would go to 0....
    Your observations are correct. The neural net is not an "AI" that learns. "Learning" happens when the neural net is trained to derive the weights and bais values for the ann model. Once those weights are locked in, it will always output the same calculated VE based on the inputs of rpm, pratio, cam position. Also, that VE is further manipulated by various multipliers afterwards, some of which we can control in the tables that are available.

    It takes a bit of knowledge in the field of data science to know what a feed forward neural net actually is and is not but if you have that knowledge and some programming skills, you can come up with your own weights and bias that can achieve a more desirable effect. The process is tedious though, requires a lot of data prep and coding. At least it used to be tedious before HPT came out with their neural net trainer which is supposed to dumb down the process a bit. I haven't tried HPT's trainer since I can do it for free myself programming with Python or R. In my experience, it's not usually worth the time but is fun for kicks and giggles when bored.

  12. #12
    Tuner in Training
    Join Date
    Jun 2017
    Location
    Richards Bay South Africa
    Posts
    23
    Hi Hemituna. Sent you a PM. Need some help on a Magnuson Ram 1500

  13. #13
    quick question can anyone tell me if this is right or not ? i swapped 5.7 dodge motor for the 6.4 in my 1500 i want to run a tune i found in repo for jeep 6.4 srt motor which is what the donor motor came out of, its asking me to purchase the tune . once i purchase the tune i should be able to flash it to my truck correct?

  14. #14
    Advanced Tuner
    Join Date
    Apr 2014
    Posts
    704
    Quote Originally Posted by grandprixdude420 View Post
    quick question can anyone tell me if this is right or not ? i swapped 5.7 dodge motor for the 6.4 in my 1500 i want to run a tune i found in repo for jeep 6.4 srt motor which is what the donor motor came out of, its asking me to purchase the tune . once i purchase the tune i should be able to flash it to my truck correct?
    No, do not flash an OS from a different vehicle to your own vehicle. There are far too many background configurations that aren't able to be edited via HPTuners for it to function properly in your truck.

    Simply open your tune file from your Ram, then open the Jeep tune file as a "compare" file, and copy over as many differences from the Jeep file to the Ram file as make sense. Not everything should be copied over.

    One example of something you shouldn't change from one file to the other is in the [Engine] [General] tab, it will have 2 fields labeled "engine size/type". One of those fields will tell you at the bottom of the window as you hover over it that "changing this value to anything other than the original value will disable esp/traction control". So, naturally you wouldn't want to change that value if you wish to keep those functions in tact.

  15. #15
    Tuner in Training
    Join Date
    Oct 2020
    Location
    NJ
    Posts
    11
    I changed the engine size from 5.7 to 7.0 to compensate for increased displacement. This seemed to help with fine tuning the VE tables. I also have NN disabled and sensed MAP enabled (cam and FI) so not sure if you are running stock or not. I left engine type and displacement stock at 5.7 as those (from what I read) can disable TC..
    Last edited by VooDooButter; 11-16-2020 at 07:33 AM.