Page 5 of 13
Posted: Wed Sep 07, 2005 2:15 pm
by efahl
whittlebeast wrote:Is that possible that that is the correct answer. Or is that a missfire feading raw fuel into the exhaust confusing the 02 ....
Well, if you read the manual... Oh, yeah, there isn't one. Anyhow, it would say something to that effect:
Beware that if you go so rich that the spark is blown out, the lambda sensor will read the unburned mixture as dead lean (what with all that oxygen still present) and the closed loop algorithm will attempt to add fuel, which the autotune algorithm will detect and adjust the VE bin richer, thus making a bad situation worse. Use with caution, this thing will not perform magic. (I'm not even convinced it will do anything useful.
)
Eric
Posted: Fri Sep 09, 2005 10:20 am
by efahl
I've been playing with some things to enhance this thing and make it more safe. The variable used to define correction will be moved into the ini file, probably something like
Code: Select all
[AutoTune]
table = veTable1Map
corrector = egoCorrection ; Variable giving percent correction to drive autotuning.
I've also been playing with a parameter to limit how far a vertex can be adjusted, which I'm calling "lumpiness."
Code: Select all
[AutoTune]
table = veTable1Map
lumpiness = 5 ; percent
This will keep the vertex that is being tuned within 5% of the extremes of the four surrounding vertices, under the assumption that the table should be somewhat smooth and will be roughly tuned to start. Note that the limits are transient, so if autotune changes vertex A by +5%, then moves to adjacent B, which also goes +5%, then the limit when we return to A will now be 5% above B.
I will also be setting the default parameters to much more conservative values than the current defaults, so that unwitting use will have a lower probability of doing harm.
Comments welcome while this stuff is still in the oven.
Eric
Posted: Fri Sep 09, 2005 2:01 pm
by MiseryQ
Great work. I really like the lumpiness factor.
Is there an easy way to keep it from autotuning below the lowest ve bin?
I think that could be the reason for the decel leaning, even though it's slight.
Posted: Fri Sep 09, 2005 2:09 pm
by Mike_Robert
MiseryQ wrote:Great work. I really like the lumpiness factor.
Is there an easy way to keep it from autotuning below the lowest ve bin?
I think that could be the reason for the decel leaning, even though it's slight.
I think that's handled by the "Y" parm:
Code:
[AutoTune] ; Users should copy this section to custom.ini to modify it.
table = veTable1Map ; Should be the map3d_id of a TableEditor entry.
; Vertex tolerance parameters
xRadius = 300
yRadius = 10
; Tuning block parameters
xLimits = 1000, 8000
yLimits = 20, 110
; Controller parameters
initialStartupInterval = 1.0
updateInterval = 1.0
proportionalGain = 0.5
"X" is RPM in all the ini files I've seen, although someone could have swapped them. Y is normally MAP, TP or load. The limit parameters box the tuning points, no changes will be made outside the box. The radius parameters dictate how close the operating point must be to the tuning point for autotune to operate, if the operating point falls outside this smaller box, then no action is taken.
Sayeth Eric.
-Mike
Posted: Sat Sep 10, 2005 8:35 am
by efahl
Above discussed parameters have been implemented in this release.
http://www.not2fast.com/megasquirt/mt/p ... _setup.exe
Note: The "allowAutoTune" setting has moved from the [TableEditor] section to the [AutoTune] section, as seen below.
Here is the complete list of settings (this is from megasquirt-I.ini), with their default values. Many more diagnostic messages are now produced in the tuning dialog, allowing you to monitor your settings...
Code: Select all
;-------------------------------------------------------------------------------
;-- Users should copy this section to custom.ini to modify it. ---------------
[AutoTune]
table = veTableMap ; Should be the map3d_id of a TableEditor entry.
allowAutoTune = on
corrector = egoCorrection
; Vertex tolerance parameters
xRadius = 200 ; rpm (probably)
yRadius = 7 ; map (probably)
; Tuning block parameters
xLimits = 1500, 4000 ; rpm (probably)
yLimits = 60, 90 ; map (probably)
zLimits = 10, 200 ; VE (probably)
; Controller parameters
initialStartupInterval = 1.0 ; seconds
updateInterval = 1.0 ; seconds
proportionalGain = 0.5
lumpiness = 5 ; percent
Posted: Sat Sep 10, 2005 8:42 am
by efahl
MiseryQ wrote:Is there an easy way to keep it from autotuning below the lowest ve bin?
I think that could be the reason for the decel leaning, even though it's slight.
Now is this a function of tuning outside the reasonable range (i.e., below idle MAP) or would a global set of Z-limits be better? I like the idea of global Z constraints, so let me work on that...
Eric
Posted: Sat Sep 10, 2005 9:11 am
by efahl
Ok, just added the zLimits (and editted the above post to reflect the ini settings). Fixing all the ini files took 10 times longer than writing the code...
Posted: Sat Sep 10, 2005 4:03 pm
by poundinfire
I just want to make sure I have my EGO settings correct to autotune to an AFR table. I am using MSNS-e 25n9.
Under "Exhaust Gas Settings"
EGO Sensor Type : Wideband
EGO Switch Point : 2.352
Ignition Events pe Step : 4
Controller Step Size : 5
Controller Authority :75
Active Above Coolant Temp : 71*C
Active Above RPM : 1300
Under "Lambda AFR Setting"
Change EGO Limit above (kpa) : 255
Change EGO Limit (+-) : 5
For VE Table 1 : ON
For VE Table 2 : Off*^(DT)
Control Algorithm for AFR tables : Speed Density
Use Target Tables : Always
Posted: Sat Sep 10, 2005 4:19 pm
by efahl
poundinfire wrote:Ignition Events pe Step : 4
Controller Step Size : 5
Controller Authority :75
I would raise that first one to the 10-20 range, and drop the second two to 1% and 15%, respectively. This will slow changes down and make things much more stable so the autotune code doesn't overshoot and cause problems.
Eric
Posted: Mon Sep 12, 2005 12:41 pm
by whittlebeast
Last night I was messing with Extra 025v2 and MT 619 I was not getting the option to Autotune
AW