Results 1 to 20 of 20

Thread: Neural network trainer

  1. #1
    Tuner in Training
    Join Date
    Aug 2021
    Posts
    36

    Neural network trainer

    I decided to go ahead and try out the neural network trainer. So far it doesn't seem too bad to use. Just have a question and it may be stupid.

    Let's say for example my exhaust cam timing is 125 at low rpm and I'm trying to tune that area with the trainer. Now in the trainer cam positions have 120 and a 130 exhaust cam position and the corresponding intake position. Which table would I need to work on to tune the 125 position? Would I apply fuel trim data to the 120 table or 130 table in the trainer?

    I have the scanner setup with filters for the different cam positions already. Thank you

  2. #2
    Advanced Tuner
    Join Date
    Jan 2022
    Posts
    882
    what does your 'normal' table look like as far as exhaust centerline?

    if it's anything like mine NN woulda already cut out by 120 deg, wouldn't it have?
    Attached Images Attached Images
    [email protected]

    don tanklage

  3. #3
    Mobile site sux on my phone; not typing it again haha.

    Screenshot_2023-01-24-12-43-55.png

  4. #4
    Tuner in Training
    Join Date
    Aug 2021
    Posts
    36
    Thank you guys! I agree mobile site sucks lol I keep switching it to desktop site to see pictures.

    As for my cam tables they have been changed from stock. It's a 5.7 truck hemi and I swapped in the 392 cam. The fuel trims were pretty far off. In the table you posted it has some low numbers in the lower rpm area and on mine it's so sluggish like that.

    In my logs idle it targets 128 and actual is about 127-128 and around 1000 or so it pulls to 125ish. Above 3k it gets to 122 or so and gets to 108 by 6k. Wot and max airflow is similar to the pt tables. Pulls strong throughout.

    Did some tuning with setting the entire cam tables to each break point and logging fuel trim (math) and got it fairly close. Seems like with nn trainer it's not perfect. I had it between 1-5% and went back to my cam timing and I had a few 9s and 10s...

    I'm thinking about trying to turn off nn and tune ve to see if I can get it better.

  5. #5
    ~5% or less sounds pretty good. You may already know this, but the high and low pressure ratio regions may be "contaminated" with some transient data (like sudden changes of the gas pedal, or dfco sending you into open loop, cat overtemp etc.). Also, cells with little data will usually be skewed somewhat.

    When I get home tonight I'll post up a current log of mine (aftermarket cam tuned with NNT) just to compare.

  6. #6
    Tuner in Training
    Join Date
    Aug 2021
    Posts
    36
    Is there a way to filter that data out so it's not skewed? The majority of them are within 0-5%.. Haven't done wot yet.

    I previously use the ipw/fuel mass way of tuning it and had big afr swings in wot. 11.2 at some points 12.7 at others. Couldn't really get it right..

  7. #7
    Tuner in Training
    Join Date
    Aug 2021
    Posts
    36
    Would you mind if I send you a message with a log and the tune that I've done so far?

  8. #8
    Quote Originally Posted by Josh19881 View Post
    Is there a way to filter that data out so it's not skewed? The majority of them are within 0-5%.. Haven't done wot yet.

    I previously use the ipw/fuel mass way of tuning it and had big afr swings in wot. 11.2 at some points 12.7 at others. Couldn't really get it right..
    When the fuel trims are fairly low like yours, you'll want to only add half of the error to your VE tables. That will help avoid overshooting the corrections.

    I've attached a file of my current graphs, with lots of filters included. Hopefully that will give you some ideas. The VE table filters result in some cells that are a few percent different than the unfiltered tables. I've found that it's still best to take your problem areas, look for them in the "chart vs. time" plots, and try to figure out if the errors should be considered in your VE table correction.

    One math in particular that saved me a bunch of time: I call it "PE Adder" - Power Enrichment Adder. Plotting PE Adder vs. time helps pinpoint any event where the PCM is not commanding 14.7 AFR; helps find all the WOT runs, COT conditions, and startup/warmup fuel additions from the FA Cold Enrich table.

    The PE Adder formula I use is simple: 1/[commanded AFR]-(1/14.7)

    Also attached is a current log of mine, trying to dial in the startup AFR, idle (with minimal chop, trying to be sneaky), and transient fuel & throttle enrichment / enleanment. You can see while cruising and accelerating smoothly that as the cam moves around, the fuel trims stay pretty neutral.

    2023-01-23 NNT Tune 559 Run 124.hpl

    2023-01-25.Graphs.xml

    2023-01-23.Charts.xml
    Last edited by rockystock; 01-25-2023 at 09:44 PM.

  9. #9
    Advanced Tuner
    Join Date
    Jan 2022
    Posts
    882
    Quote Originally Posted by Josh19881 View Post
    As for my cam tables they have been changed from stock. It's a 5.7 truck hemi and I swapped in the 392 cam. The fuel trims were pretty far off. In the table you posted it has some low numbers in the lower rpm area and on mine it's so sluggish like that.
    I bet it is. Rocky has it backwards, as you retard the cam the exhaust centerline numbers get smaller, not bigger. 392 cam has more duration than your 5.7 cam did (intake valve closes later when at same centerline). Longer duration likes static compression increase, not decrease. Arnt 5.7's 9.5 to 1? 392 is almost 11:1 . I am sure it will make more power up top but everything is a trade off.

    *that ^^^ was in reference to the phone pic above, not to what was just posted
    Last edited by LilSick; 01-25-2023 at 09:53 PM.
    [email protected]

    don tanklage

  10. #10
    Quote Originally Posted by LilSick View Post
    Rocky has it backwards, as you retard the cam the exhaust centerline numbers get smaller, not bigger.
    Huh? Where did I say the opposite?

  11. #11
    Advanced Tuner
    Join Date
    Jan 2022
    Posts
    882
    Oh my bad. You said the cam is max advanced at low rpm. 392 tune has it retarded at low rpm. It's all good, glad he has someone to help him. Bunch of tight lip mf in dodge community. Would think its pro stock or something
    [email protected]

    don tanklage

  12. #12
    I was pretty sure all the OEM setups go max advance below about 1000 RPM. The RPM Enable table may allow for adjusting the low-RPM position. It's been a minute but I thought I tried zeroing out the RPM Enable just to tune the different cam position VE tables down low; no matter what I did the cam always went to full advance near idle. There's probably a hidden parameter that's unavailable in the calibration...

  13. #13
    Advanced Tuner
    Join Date
    Jan 2022
    Posts
    882
    Quote Originally Posted by rockystock View Post
    ...no matter what I did the cam always went to full advance near idle.

    There's probably a hidden parameter that's unavailable in the calibration...
    That alone could save someone a bunch of time.

    Thx
    [email protected]

    don tanklage

  14. #14
    Josh - I took a look at your log and tune. Fuel trims look good to me - just a little lean around 1200-1400 RPM but not bad at all (wish my afr was as steady lol). Your WOT / open loop runs are fairly well dialed in as well - around 2-3% leaner than commanded from about 3800-4700 RPM.

    I've usually had best results with the NNT by getting good data over as much of the tables as possible, and uncheck the 20 unused tables so they're not used for training.

    Looks like your Vacuum Mult table (ECM [34082]) was disabled by adding 400 kPa to the axis values. Your table is probably adding about 1.3% to your fuel mass, because I think the lowest cell is being applied all the time. Not a big deal though.

  15. #15
    Tuner in Training
    Join Date
    Aug 2021
    Posts
    36
    Quote Originally Posted by rockystock View Post
    Josh - I took a look at your log and tune. Fuel trims look good to me - just a little lean around 1200-1400 RPM but not bad at all (wish my afr was as steady lol). Your WOT / open loop runs are fairly well dialed in as well - around 2-3% leaner than commanded from about 3800-4700 RPM.

    I've usually had best results with the NNT by getting good data over as much of the tables as possible, and uncheck the 20 unused tables so they're not used for training.

    Looks like your Vacuum Mult table (ECM [34082]) was disabled by adding 400 kPa to the axis values. Your table is probably adding about 1.3% to your fuel mass, because I think the lowest cell is being applied all the time. Not a big deal though.
    That's odd. I never touched the vacuum mult table. I just compared to the stock tune and it's untouched. Should I adjust that table? Wasn't too sure on the affect so didn't modify it.

    Also I did decide to turn off neural network and fuel trims are a bit more stable, wot is more stable. It overall just feels better. I think we have it dialed in pretty close with and without the neural network. So if we run into any problems I can go back to NN on and go from there. Thank you everyone for the advice and help! I really appreciate everything.

  16. #16
    Tuner in Training
    Join Date
    Feb 2007
    Posts
    11
    Quote Originally Posted by rockystock View Post
    I've usually had best results with the NNT by getting good data over as much of the tables as possible, and uncheck the 20 unused tables so they're not used for training.
    How do you determine which tables aren't used?

  17. #17
    I tried to explain in a couple earlier threads.

    Hemi - one cam, exhaust and intake lobes - fixed lobe separation. If you choose your max and min cam training values so the sum of the max exhaust and min intake equal the total lobe separation (in crankshaft degrees), and the sum of the min exhaust and max intake equal the total lobe separation, then you will have 5 tables in the NNT where the sum of the exhaust and intake angles equal your total lobe separation. Use those tables and don't use the others.

  18. #18
    Advanced Tuner
    Join Date
    Jan 2022
    Posts
    882
    What's the difference between lobe separation and total lobe separation?

    Or is that just what you call it when speaking in crank degrees for something that is measured in cam degrees? (on a cam card)
    [email protected]

    don tanklage

  19. #19
    I believe all relevant PCM parameters are in crank degrees. And for tuning, the PCM will only see a fixed value for "fully advanced" regardless of the actual cam specs. For example, my 2010 Challenger NGC4 PCM shows 127 degrees (exhaust angle) at full advance, regardless of whether the engine has the OEM 5.7 cam, a stock 392 cam, or an aftermarket. Likewise, the lobe separation is fixed at 228.5 crank degrees, despite my FRP Mark 82 cam with its 226 degrees (I think).

  20. #20
    Tuner in Training
    Join Date
    Feb 2007
    Posts
    11
    Quote Originally Posted by rockystock View Post
    Hemi - one cam, exhaust and intake lobes - fixed lobe separation. If you choose your max and min cam training values so the sum of the max exhaust and min intake equal the total lobe separation (in crankshaft degrees), and the sum of the min exhaust and max intake equal the total lobe separation, then you will have 5 tables in the NNT where the sum of the exhaust and intake angles equal your total lobe separation. Use those tables and don't use the others.
    I know all those words were in English, but I think my current BAC is way too low to properly translate.