VEMS Community Forum
VEMS => Software => Topic started by: multiplex on April 06, 2010, 08:59:48 am
-
seems like a great feature, i'm just not sure if its generate realistic numbers?
in the old days i would use MLV to tune my map and re-import into vems.
with vemstune, that doesn't work all that well
any tips on using tune by statisics? or is there still work to be done
-
I´ve tested it from a log. Looked to make the right adjustments.
Just give it a go and see if it comes up with sensible numbers.
-
seems like a great feature, i'm just not sure if its generate realistic numbers?
in the old days i would use MLV to tune my map and re-import into vems.
with vemstune, that doesn't work all that well
any tips on using tune by statisics? or is there still work to be done
for me works very good if rpms/kpa dots are set good.
-
for me works very good if rpms/kpa dots are set good.
What you mean with dots?
Do you used MLV before for analyze? How it compares?
-
for me works very good if rpms/kpa dots are set good.
What you mean with dots?
Do you used MLV before for analyze? How it compares?
I used mlv, and vemstune is better.
My engine has 180Nm at 2000rpm, 168Nm at 2200rpm, 183Nm at 2400rpm. So if i put rpm dots 2000 2400 tune by statistics doesn't work well, but i i put rpms dots 2000 2200 and 2400(because engine here has some resonance) tune by statistics works great.
-
Clear! I will try it shortly! :)
One feature I miss since MSTweeks3000 times: this software allowed plot datalog in colored plot. So it was possible to determine optimal places for rpm bins/dots very easy...
-
What you mean with dots?
He means the "first differential of" (RPM or MAP) or "change of"
Not had much success with this myself, but that is a lot to do with the crazy cam I have - the statistics don't know which samples to bin. My own stats tuning engine is broadly similar but I have (by necessity) had to put a lot of effort in to throwing away the samples where horrible things were happening... This means I get better results.
-
dnb, how you compare results of both: MLV and VT analyser?
-
I haven't used MLV, only the VT analyser vs my own analyser. (Obviously mine is better, but I'm biassed and it's been heavily optimised for big cammed v8 engines...)
-
I did my map with vemstune, maximum error is 2% :). But i have data from 300km trip for tuning.
-
I datalogged to the internal SD card and downloaded 2x90 mins log of me driving 250 km and optimized my cruising VE bins to hold +/- 1% close to target, worked very well.
Be sure to have rpm and pressure bins where the engine spends the bulk of the time. This means that you keep an eye on the rpm at which you hold certain legal road speeds, 70-90-110 km/h are normal here in Sweden. Those rpm bins will then be very easy to optimize since the data will hit those bins very often.
Don't make changes to bins where you have a small number of datalogged points and where the data suggests large correction, this is usually due to engine break (fuel cut) or acceleration enrichment.
-
Does analyser filter out records after unstable situations like acceleration enrichment?
-
I'm not clear which log file and VE map gets used for this analysis tool. For the VE map, I just open a saved config file. But for the log file that matches that config, how do you choose it? There seems to be no option on the VE Tune by statistics page. Is it just the newest one?
-
A log file saved with VemsTune contains the configfile and therefore the VE map is known, so the analysis will suggest alterations to that VE map. The log also contains configuration changes, I'm not sure if/how that is taken into account when the analysis is made. My point being, don't make any changes to your VE or lambda target map while datalogging if you wish to do analysis later. Do it in steps. Datalog, analysis, apply changes, .. etc.
The SD card logs do NOT contain the config, so you need to load a save config of the one used when the SD card log was made if you want to do any analysis and alter your VE map. VemsTune makes a point out of this when you open a SD card log.
As with other things in VemsTune this is a moving target, bear that in mind and suggest stuff and ask away if the function is not clear or needs enhancements.
-
Thanks. I'm off to do some datalogging and tuning then.
-
In the next version, some changes are started to the tune by statistics
now removes outliers in the data, adjustable of course
next step is that it will draw a bell curve showing your data spread
and after that, weighting according to how close to the actual site your are will be added.
-
sounds like some awesome features in the works
i have one question - how does running map multiplication effect the ve stats? Does it need to take this into consideration when calculating the results?
Also, how can i save changes i make to the filter? I see a button there, but its always grayed out. I usually filter out my idle area so that isn't effected
-
You can select the cells to "fix" . so that not all areas are changed even if you let the check go through them.
Running map multiplication will not change anything.
The VE map works the same either way, i.e 10% more in the VE table adds 10% to the final pulsewidth.
-
Any recommendations on the values to enter for lambda delay, variance multiplier and min # data in cell?
-
And how can I close out the transients from statistics?
-
The log you do is the bases for any good tuning.
so it´s best to do a specific driving, like alot of steady state cruising and such in various gears.
in-town traffic ain´t exactly going to help the tune get better.
If you can keep it at 2000rpm steady in a few gears, then same for 3k and 4k, maybe 3-5sec per rpm and gear so that everything stabilizes. Then a WOT run if the mixture is close enough.
-
The results I have had with tune by statistics I personally had were not as good in comparison to MLV using the very hard option.
Tune by statistic leaned out a lot of my figures for some strange reason, I'll test it in the next vemstune version
-
To use the tune by statistics should the EGO limits be set to 0 for the logs?
-
To use the tune by statistics should the EGO limits be set to 0 for the logs?
no
-
The results I have had with tune by statistics I personally had were not as good in comparison to MLV using the very hard option.
Tune by statistic leaned out a lot of my figures for some strange reason, I'll test it in the next vemstune version
newer versions have been a lot better, I'll have to do a side by side comparison of MLV on my next logging run
-
Does tune by statistics not work if you are running alphaN?
The axis are MAP x Engine speed, so all my data is at 100kpa (as I don't have the MAP attached to the throttle bodies
Although I have just spotted that MAP(load) multiplication has switched itself on, even though it was definately off in an earlier config ???
I'll turn it back off and try some more logging.
-
It should work just fine with AlphaN, its only interested in matching the tables as they stand with the lambda delta between target and reading.
-
I'll try it again tomorrow now I have switched off MAP (load) multiplication which switched itself on (maybe at the same time as Wideband changed to narrowband)
However the log I took today came up with this
(http://www.fautley.co.uk/copper/albums/userpics/normal_vetune.JPG) (http://www.fautley.co.uk/copper/albums/userpics/vetune.JPG)
-
What am I doing wrong? Why is VEMSTUNE tune by stats giving me such radical numbers in places (if I didn't filter out the off-throttle lean figures, some of the +s were in the 20s!!! That would drown my engine (I know from my last track outing and some figures I had causing rich misfires):
(I know the map is out of whack, this was in the middle of tuning it)
http://i.imgur.com/mdkrs.png
(http://i.imgur.com/mdkrs.png)
-
Did you read my posts earlier in this thread?
Only adjust cells which have low min/max figures and low variance, that usually also means a large number of hits. Your job as a tuner then is to realise that the neighbouring cells, even if you have not visited them, are just about the same. This is a great tool if used properly. Overall, smooth out your VE table - having it look like that is only going to make any "auto tune" tool have a difficult time getting it right.
Copy the "104" kPa line all the way to the top. Press and hold ctrl on each cell, then arrow up to copy them to cells above it while moving up.
-
Thanks Mattias. The upper cells aren't in use yet as I'm trying to get the engine working as well as I can naturally aspirated.
The problem is that during the tuning session, I only seemed to hit some cells under fuel cut. So it would suggest I richen it by 38 points with a variance of 1.7. It's not until I clamp the lambda value that it starts being closer to MLV but there are still those 8s and 10s. I guess what I'm really getting at is, why can't VT take all the things into account that MLV seems to, or am I just using it wrong?
Also, is it possible to add some intelligence to the way VEMS works with lambda values? I had a few misfires due to being too rich. I have an AEM UEGO installed for my AFR gauge (and also as a backup sensor/info source) and when this happens, it seems to ignore the ultra lean values if the trend before that was to super rich - it pegs at 10.0, it's min reading, until it gets a reading in its range, while VEMS reads 18.9 and tries to correct as such.
-
Let say you have a "50" in one cell, the numbers above and below that in the rpm column will be very close to that.
So you should be able to look at the logs yourself and figure out the change required, always after changing two values smooth out the map so that between cells you are sure are right.
for instance if you hit dead on at 4500rpm and 80kpa then you´ll know that the value there will be very close to right.
It shouldn´t take more then 30min of driving to get a fuel map 95% there. It´s also very much practice.
I don´t feel you should be using auto tune unless the map is 95% there and your just trying to get it very accurate.
Also as mentioned don´t trust any software when it´s suggesting extreme amounts, just calculate it yourself or do an estimate based
on the values in nearby cells.
-
^^^
as above, rough out the VE map manualy using the logs where there are large differences, then smooth out the map in 3D, then use the auto tune to fine tune the map toi get you to around +-5%. afterwards set rich and lean limits to 5%
If you are getting rich missfire, you really need to manualy tune the map in that area acordingly.
Alternatively use a dyno and use the 3D tuning facility and hot keys. Take about 30min to get the map done everywhere
-
Thanks guys. Yeah I learned that the hard way unfortunately. :) I went out the first time after roughing in the map and doing some quick tests around the block. I can't drive it around too long because it's loud and, well obviously not street legal. I went on the track and it was quite good except for one hiccup. I come back to the pits, run it through MLV and it tells me to add a pile of fuel in a few places - I thought I was almost melting my engine or something with such a lean area. But it was a misfire fooling it into thinking it's ultra lean. Only after going out again and realizing it's worse, and seeing my AEM show 10s I started changing it by hand.
Now I turned ego down to 1%, turned up the EGT correction (which was at 650c and 20% fuel by default - my ecu was determined to drown my engine lol), now I think I have a map that's close to reality (the one in the earlier post was an intermediate, I've evened out the values). Time for more track time!
It's my first standalone experience, it's been a ton of fun but the learning curve is humbling.
-
Hi guys, I am a bit frustrated about Vemstune stability. I've tried last stable version (09-07), Tuning by statistic - I made log, then (with connected ecu) run Tuning by statistic - crash. I had converted log to CSV, run MLV, tune VE table and then want to import VE table back to Vemstune. The 1st import was ok, then 2nd attempt no - both VE axis were corrupted - VT restart didn't help. I really don't know how to use VT for VE tuning in praxis. Another thing is how terribly slow is VT Log viewer (I have Dell D430 2x1.33Ghz Core2Duo, 2GB ram, it's not enough ???) it's soo laggy, (I can make some video record and show it).It's nice to have antialiased lines, but more important is speed and usability IMHO. MLV is absolutely without any problems and very fast. Anyway VT is still improving, I know it, but I think, the VE tuning/logging capabilities should be on the first place..... Thanks.
-
Vemstune can export msq file in right format just once. Any further export is corrupted: MLV reads it, but VT - not. Then simply restart Vestune and you can do it again.
Developers wrote it is solved. Probably included in last nightly version.
Anything better with tuning by statistics? Is there some improvements since beggininig of year?
Gints.