History of GenBoard/Manual/VeLearn/Testing
Older Newer
2005-06-22 06:39:36 . . . . MembersPage/MarcellGal [how to make VE learning faster]
2004-12-06 07:26:45 . . . . MembersPage/MarcellGal [started page]


Changes by last author:

Added:
The learnt values are not saved automatically. They can be

* viewed on LCD (mtt table ... ?)

* or saved to the reference (j[] table) with mvs command

* or saved to PC with mct command

so they can be examined before using further.

----

There are some nonstandard experiments on

MembersPage/Gabor (fully mappable NBO2 target was dropped when WBO2 became fully operational. Maybe we can add it back - using l table for NBO2 voltage target as it was before. Even though using NBO2 closed loop at higher power loadsites is risky, some NA engines that tried it didn't blow up yet).

Some compile time settings, original values:

* define VE_LEARN_SAMPLES 16 and

* define VE_LEARN_SKIP_SAMPLES 10

was decreased by MembersPage/Gabor (appr to 25%) to successfully increase VE learn speed and the learning was still convergent (learning algorithm is same for WBO2 and NBO2, just WBO2 input is more precise and fast over the entire range). True as it is, the above original values were set rather too high, very safe values. If you decrease them under 50% of original, watch the tables often (eg. in MegaTune close the 3D view, and reopen it: theoretically it reads the new learnt values from GenBoard)

We usually disable VE learning after tuning, since there is not much need for it. The ego correction will show anyway if some loadsites (or warmup enrichment settings, or whatever) are way off.

----

See also

* GenBoard/Manual