[s-cars] Fuzz logic vs adaptive learning

[QSHIPQ at aol.com]

I understand the differentiation between FL and AL as the following...  
Adaptive learning (AVC-R) means you have a set of parameters you want.  The 
controller applies those parameters and says no or yes to them as presented.  So the 
map is presented based on the exact number of variables you can program, it's 
map is entirely based on exactly that number of inputs.  "Adaptive" means it 
does have the ability to give less or more of a given parameter, but can't fill 
in an unassigned value.

Fuzzy Logic (HKS EVC-IV, Profec A) learning means you have a set of paramters 
you want, and FL will say yes or no and come back with an acceptable 
parameter based on desires and other inputs.  FL is capable of assigning values to a 
map far in excess of what you may be able to program.  In other words, you give 
it guidance and it can fill in anassigned values to get you there.

The biggest risk to adaptive and FL controllers IMO, is exactly where they 
have the highest use, CO.  Either can be programmed at a specific altitude and 
compressor efficiency, but both controllers will create a bad engine 
environment as soon as altitude/press changes affect turbo CE and engine VE.  In other 
words,  AL/FL controllers aren't desireable in an environment that has a high 
variance from "standard atmosphere"  (point in time programming).  They also 
require that ANY change to the hardware (engine tweeks that affect VE or CE) 
require a "relearn/reprogram".

As such, it's my claim that the best controller is still motronic.  Since 
AVC-r has no input to MAF it's not a Mass of air based (motronic) profile.  And 
in order for it to be a true speed density system, it requires feedback loop 
with MAP (read it has no O2 input).   Motronic has MAF, MAP and O2, hence can be 
either a MAF or speed density computer (yup Minhea I won't go there:)   Which 
means that any significant change in altitude will not be a recognized input 
to the boost profile of the external controller.  I see that as bad, nee, 
potentially really expensive.

WRT motronic, it does have some AL built into it, since it can learn that you 
are at altitude and change the baseline fuel on startup (hi alt, low temp) 
and can enrich fuel mixture if knock is detected.  Knock control itself is a 
"learning" mode, and is adaptive in the 20vt (fuel and driving conditions) and 
has really been around since the mac 11 (premium vs regular fuel tables).

Integration of boost control with engine management is crucial, and is really 
what gives Motronic the massive edge over a standalone controller.  HKS has 
made several attempts over the years to integrate boost control with other 
controllers, but has pulled many off the market because of the complexity of 
integration without direct interface with the engine management computer.  Profec A 
(fuzzy logic controller) has been discontinued, I suspect from reading the 
other boards, from the exact problem above.   My thinking is, that makes any 
argument that "software" in a controller is better or can be better than 
motronic:  moot.  From what I've read to date in terms of "performance" clearly 
indicates to me that the WGFV in the motronic is a problem.  Once that hardware has 
been addressed, I really believe the software already has all the makings for 
one of the best boost controllers on the market today.  AND you already own 
the darn thing.

To make the claim an external boost controller is better than motronics boost 
controller, it needs minimally the same inputs as motronic.  Then it's still 
not interfaced with the ECU like motronic boost control.  Then, if you finally 
get the controller to have the same inputs and interfaced control, you either 
have a redundant set of sensors, or you have just built a better motronic 
box.  We already know a lot about that box under the passenger foot.  Why not 
just make it better at it's job?

Scott "more FL" Justusson
QSHIPQ Performance Tuning


In a message dated 1/23/2004 6:13:22 PM Central Standard Time, 
vfregeac at sympatico.ca writes:
Working with the delta between expected and actual is the definition of
"closed loop" controllers, which is any controller with a sensor for
what it controls. "Fuzzy logic" does a bit more. It react to this gap
once and "remember" what happens so next time, instead of trying
different parameters, it will apply what it learned from the last time.
Moreover, if it learns from two set of conditions, it can deduct what to
do in between these conditions. In fact, in any application when you
can't program all the possible conditions, a learning "fuzzy logic"
controller is superior to a classical binary logic programmed
controller.

BTW, I'm thinking that, as we have to reset the Motronic to erase what
it has learned, there is a good chance the Motronic includes a "fuzzy
logic" part for this "learning" process.

My 0.02 cents of experience in fuzzy logic and other controllers, no
BTDT in engine control though.