Does the Ducati ECU 'learn' from itself?

[Charlie98]

Speaking only of more modern Ducatis, and particularly something like my '13 796 with the Siemens ECU...

I was reading some comments about an exhaust swap and such on FB the other day... and the question always comes up:  'Do I need to reflash the ECU?'  Someone later commented that the ECU was 'self-learning' and would adjust to, or compensate, for the new exhaust and the increased airflow through the motor.

I understand the function of the ECU, particularly with all the sensors it has these days, is to make tiny adjustments to the fuel, air, and ignition... but does it really accommodate something like a slip-on exhaust?  I have slips on my 796, they improved the performance of the motor, but that was from removing the restrictive catalyst-filled exhaust cans... not from the ECU adjusting so much.  It still pops on roll-off, and I'm sure it's running lean, because it basically came that way from the factory.

Dunno... I'm just curious if 'self-learning' is a true statement, and to what extent?

[stopintime]

Not learning, but to a degree reacting.

My impression is that it does not react enough.

[Orange16]

Any of the closed loop ECU have an adaption process, where ongoing trims made to the fuelling via the narrow band lambda sensor feedback loop are stored and then used for future running.  It's a continual process.

Closed loop ECU on Ducati models started with the 2006 MY - new models S2R1000, S4RS, SC1000 and updated ST3, all using the 5AM ECU and a single lambda sensor.  All new models thereafter were closed loop, whereas some of the older models - 800SSie, 1000SSie, S2R800 - stayed as open loop as they were soon to be dead.

Future ECU types used - Conti M3C in M696, 796, 1100, Melco in Panigales, MTS1200, Bosch, later Marelli etc - are generally ECU that were never developed with open loop variants in mind.  Therefore, turning the lambda sensors off is not a simple yes/no process like it is with the Marelli 5AM.

Closed loop simply means that the output of the narrow band lambda sensor is used to adjust the mixture.  The narrow band lambda sensor is a disimilar metal galvanic cell that outputs a voltage in the absence of oxygen, and that output is realistically a yes/no switch output in usefulness.  There are wide band sensors that will give you a reading of air/fuel ratio over a wide range rich to lean.  The narrow band sensors fitted OEM do not do that.

The narrow band lambda sensor output is theoretically 1V or 0V, but in practice it is 0.8-0.9V for an air/fuel ratio richer than 14.7:1, and 0.1-0.2V for an air/fuel ratio leaner than 14.7:1.  In the lambda scale, 14.7:1 (or around that, the stoichiermetric ratio varies for different gasoline types) is Lambda 1, richer is 0.99 or less and leaner is 1.01 or more.

So we'll say the lambda sensor gives an output of 0.9V for rich and 0.1V for lean.  When the engine is operating at steady state - RPM and throttle opening are constant or as close to constant as the ECU requires - and within the specified scenario - as defined by the emissions regulations being worked to, Euro 4 is anything under 130km/h I believe - the ECU will look at the lambda sensor output and then use that to adjust the fuelling to cycle the mixture around the 14.7:1 target.  That's because the cat convertors like a mixture cycling around 14.7:1 for best operation.

If the lambda sensor is putting out 0.9V, the ECU will reduce the fuel pulse widths in defined steps until the lambda sensor output drops to 0.1V.  Then the ECU will increase the fuel pulse width again until the lambda sensor output increases to 0.9V again.  And continues to do so for as long as required.  If you look at the lambda sensor output voltage of an engine idling in closed loop, for instance, the value oscillates in a very sine wave pattern between 0.9 to 0.1V.

How much change the ECU makes to the fuel pulse width is stored by the ECU in what is commonly called the adaption table, and that table is used in future fuel pulse width calculations.  It is also an allowance for manufacturing tolerances, engine and component wear, etc.

It doesn't work in transient conditions, such as acceleration, as you need repeated samples at the same - or within the range defined -RPM and load data points.

In terms of relearning after changes made - ie, I've fitted an aftermarket exhaust and the manufacturer of said exhaust is bullshitting people that it'll be fine as they'll sell more that way - it's pretty much useless.  The process happens only at load situations that the vehicle will see in steady use situations.  If you've ever data logged a motorcycle in commuter/city or touring situations, you'll know just how little throttle opening is used.  None of it will do anything for higher load/throttle situations.  And any learning will simply make the bike ran as crap as it did before.

Although saying that I believe some bikes - litre sport bikes, etc maybe - are now coming with wide band sensors, but I don't know how wide ranging the software is in using the info provided.  As ever, what the tech can do, and what it is allowed to do via Gov't regulations, are two very different things.

And flashing the ECU to be open loop and therefore ignore the lambda sensor output and associated ECU logic will usually make the bike run much nicer at lower speeds, stop them back firing on over run, etc.