Auto Tune, how does it work ?
Extreme HDF Member
Joined: Aug 2013
Posts: 16,271
Likes: 15,108
From: Somewhere SW
Extreme HDF Member
Joined: Aug 2013
Posts: 16,271
Likes: 15,108
From: Somewhere SW
My not so technical take on auto tune is as follows, at least on the Vance & Hines version. Narrow band sensors are only accurate in a narrow window where as wide band operates in a wider range, not so hard
. In the V&H literature it states that during the auto-tune segment of populating the ecu tables prior to download, the motor often tends to run raged until such time that you manually download the collected information to the ecu. My best guess is that the tuner is varying the mixture to the engine until it gets into the narrow band window that it knows is accurate, that's why it runs ruff during the learn cycle. After varying the mixture from rich to lean it sees what it believes to be an accurate narrow band reading, it records that reading to the cell that correlates to the different variables being looked at. After filling in 100 to 200 cells on the matrix it gets to the point that it can reliably predict the proper mixture for the condition sensed. I am more than likely wrong, but it's my best guess.
So if the wide band sensor can make an accurate tune in one try, the narrow band sensor may take 2 or 3 sessions to get to where the wide band is in one. If I'm wrong I wouldn't be surprised because I am making a few assumptions that I believe to be true but may not be. I am looking forward to comments from those more knowledgeable than I, so that I might be one step closer to understanding how auto tune works.
. In the V&H literature it states that during the auto-tune segment of populating the ecu tables prior to download, the motor often tends to run raged until such time that you manually download the collected information to the ecu. My best guess is that the tuner is varying the mixture to the engine until it gets into the narrow band window that it knows is accurate, that's why it runs ruff during the learn cycle. After varying the mixture from rich to lean it sees what it believes to be an accurate narrow band reading, it records that reading to the cell that correlates to the different variables being looked at. After filling in 100 to 200 cells on the matrix it gets to the point that it can reliably predict the proper mixture for the condition sensed. I am more than likely wrong, but it's my best guess. So if the wide band sensor can make an accurate tune in one try, the narrow band sensor may take 2 or 3 sessions to get to where the wide band is in one. If I'm wrong I wouldn't be surprised because I am making a few assumptions that I believe to be true but may not be. I am looking forward to comments from those more knowledgeable than I, so that I might be one step closer to understanding how auto tune works.
The thing about the wide band sensors is that they can accurately read outside of the range of the narrow band sensors.
So basically the big advantage is accurate verification from let's say 12.0-15.0 (or what ever their limits are) So if you're running 13.8 in one cell the ECU can verify and correct that cell using the sensor (closed loop) whereas going strictly off the VEs (open loop with narrow band or no sensor at all) the ECU just estimates how long to keep the injector open to achieve that 13.8 based on the value of the VE for that cell.
Mind you this is after the tune is constructed and flashed to the ECU.
Last edited by JustDave13; Nov 30, 2015 at 05:40 PM.
Road Master
Joined: Mar 2007
Posts: 1,177
Likes: 23
Seriously..... You guys are really starting to get your heads around this stuff.
With a wideband setup, you can verify the tune on the fly running at the AFR you really want to be.
The more you understand, the easier it all gets.
Andy
With a wideband setup, you can verify the tune on the fly running at the AFR you really want to be.
The more you understand, the easier it all gets.
Andy
Extreme HDF Member
Joined: Aug 2013
Posts: 16,271
Likes: 15,108
From: Somewhere SW
The one drawback is that a carb is all mechanical so changes happen instantaneously but with a tuner everything relies on data moving from one component to the other so things are acknowledged, logged and adjusted after the event takes place.
Last edited by JustDave13; Nov 30, 2015 at 07:56 PM.
Tourer
Joined: Jan 2015
Posts: 344
Likes: 4
From: USA
THIS is possibly THE BEST THREAD yet for helping me (and the other possible ECU/ECM/TUNER/ETC ludites amongst us 
) to understand what's going on with our bike's electronic/fuel/computer system. This should be a stickie!!!
) to understand what's going on with our bike's electronic/fuel/computer system. This should be a stickie!!!
Road Master
Joined: Mar 2007
Posts: 1,177
Likes: 23
We really need to talk about AFF (Adaptive Fuel Factor) and Block Learn.
If you look at the left side of this screen shot...
http://www.nbs-stl.com/HarleyTuning/...ed%20Views.png
You can see along the bottom (horizontal axis) is RPM. On the Left (vertical axis) is MAP or Load on lots of tuning software outside of the Harley World. In the field, in color, is the O2 voltage. You can see that is some places in the plot, there is lots of blue or low voltage. Other places are strongly red or high voltage. Yet other places are lots of different colors and there is no clear pattern.
On the right, I group the same data into buckets and the number is the average voltage of all the dots that fell in the box. This is known as histogram view.
Now notice the larger boxes. Those are what is known as the block learn boxes. If you have AFF turned on, the code will average all of the numbers that fall in the big box. The code stores that average as the AFF. I totally made up the boxes in that shot but every code I have seen have these boxes. They change from one code to the next. But they are there. Now notice that if your motor has both lean areas and some rich areas in the same box, the code may come up with an average 100. Even if the answer could have been very different if the box was set up in the code differently. It is your goal as the tuner to get every VE close to perfect so none of this stuff make much difference one way or the other.
In this screen shot, only the blue box on the right would make sense in the example.
Here is a shot of what this stuff looks like in a real log that needs tuning in a big way.
http://www.nbs-stl.com/HarleyTuning/...g%20Tuning.png
Both cylinders have clear patterns where the ECU is adding and pulling fuel.
Notice how the colors are showing up as boxes. Those boxes are what I was trying to demonstrate in the first screen shot.
Hope this helps explain just one more piece of this Autotune stuff make a little more sense.
Have fun tuning
Andy
If you look at the left side of this screen shot...
http://www.nbs-stl.com/HarleyTuning/...ed%20Views.png
You can see along the bottom (horizontal axis) is RPM. On the Left (vertical axis) is MAP or Load on lots of tuning software outside of the Harley World. In the field, in color, is the O2 voltage. You can see that is some places in the plot, there is lots of blue or low voltage. Other places are strongly red or high voltage. Yet other places are lots of different colors and there is no clear pattern.
On the right, I group the same data into buckets and the number is the average voltage of all the dots that fell in the box. This is known as histogram view.
Now notice the larger boxes. Those are what is known as the block learn boxes. If you have AFF turned on, the code will average all of the numbers that fall in the big box. The code stores that average as the AFF. I totally made up the boxes in that shot but every code I have seen have these boxes. They change from one code to the next. But they are there. Now notice that if your motor has both lean areas and some rich areas in the same box, the code may come up with an average 100. Even if the answer could have been very different if the box was set up in the code differently. It is your goal as the tuner to get every VE close to perfect so none of this stuff make much difference one way or the other.
In this screen shot, only the blue box on the right would make sense in the example.
Here is a shot of what this stuff looks like in a real log that needs tuning in a big way.
http://www.nbs-stl.com/HarleyTuning/...g%20Tuning.png
Both cylinders have clear patterns where the ECU is adding and pulling fuel.
Notice how the colors are showing up as boxes. Those boxes are what I was trying to demonstrate in the first screen shot.
Hope this helps explain just one more piece of this Autotune stuff make a little more sense.
Have fun tuning
Andy
Last edited by whittlebeast; Nov 30, 2015 at 08:38 PM.
Road Master
Joined: Mar 2007
Posts: 1,177
Likes: 23
Keep in mind that we are talking about the second to last line in this shot.
http://www.nbs-stl.com/MLVDemo/012%2...ngEquation.PNG
AFF is the long term part and CLI is the short term part. Harley builds the Long term lookup tables based on the patterns it sees in the short term part of the formula. What is a little unique to Harley code compared to most systems I mess with is that there is relatively few boxes in the Long term lookup table. Like 10 boxes and not hundreds of boxes.
Andy
http://www.nbs-stl.com/MLVDemo/012%2...ngEquation.PNG
AFF is the long term part and CLI is the short term part. Harley builds the Long term lookup tables based on the patterns it sees in the short term part of the formula. What is a little unique to Harley code compared to most systems I mess with is that there is relatively few boxes in the Long term lookup table. Like 10 boxes and not hundreds of boxes.
Andy
Road Master
Joined: Feb 2014
Posts: 804
Likes: 237
From: TN
Wanted to pass on some info on target tune. I had a conversation with Jamie at FM about some readings I was getting with target tune when running logs. When you auto tune with tt, you have to select the tt wideband sensors as the source, and when you run the log program you still have to use the tt wideband sensors, not the ECM source. It seems there is a difference. So I went out and did some logging, and the numbers were right where the tune was set. Andy, Jim, some of the fields that were listed are not available thru the tt sensors. That's pertains to those using the tt module only.
Warren
Warren
HD Forum Stories
The Best of Harley-Davidson for Lifelong Riders
7 Surprising Harley-Davidson Products that Are Not Motorcycles
Verdad Gallardo
8 Best Harley-Davidson Motorcycles Ever
Pouria Savadkouei
10 Worst Harley-Davidson Motorcycles Ever
Pouria Savadkouei
Killer Custom's Jail Break Is The Breakout That Refused to Blend In
Verdad Gallardo
Crazy Bunderbike Build Looks Amazing, But Is It Impossible to Ride?
Verdad Gallardo
Harley-Davidson Reveals Super Cool Cafe Racer Concept
Verdad Gallardo
Engraved Rebellion: Inside Bundnerbike's Glam Rock II
Verdad Gallardo
10 Motorcycles You Should Never Buy
Joe Kucinski
10 Things Harley-Davidson Needs to Fix in 2026
Verdad Gallardo
Thread Starter
|
Cruiser
Joined: Aug 2015
Posts: 113
Likes: 4
From: Dartmouth, Devon, UK
Just to clear up one thing, this started as a theoretical discussion not related to any particular tuner as my post #1 suggests, but thread drift has had its wicked way !
This is mainly to Andy, and thank you for the best information yet received.
This question is to Andy, here is your scatter plot, a stage one lambda map and an o2 sensor graph. Bearing in mind Harley type Auto Tuners (and other names) are sold to be used with the stock o2 sensor, and a typical map may call for a lambda of 0.750, but the o2 sensor maxes out at 0.80 (or does it?), how do these manufacturers get that AFR let alone get the Auto Tune to work ? BTW I am very conversant with a lot of closed and open loop feedback systems, but not in regard to EFI applications.
Thanks
Dave