What are your smoothing techniques

[Dirk Diggler]

and how much do you smooth by.

Also what would you conisder a dialed in car? A BEN_Factor of +/-2% or are you all going grey trying to get the to 1.00

[Black02SS]

I have recently tried smoothing by 70% and it seems to be too much but the table looks good. I don't know how we can smooth the ve table to get it to look good and still retain the +/- 2% that we need. Not unless we just need to keep going over and over on it until it falls into place.

[Dirk Diggler]

Its gotten to the point where i think its moving three digits past the decimal.


Pics of the log and tune to follow in t-5 minutes

[Black02SS]

I believe the same as I was double checking the idle on a car last night. I had approx 43 after the Ben Factor, but after I smoothed the table 70%, it jumped up to 47 and messed up the actual afr. I know that there is a reson for a smooth ve table, but how can we get it to be smooth without it altering the actual afr? I flashed this car a total of 9 times to get it all dialed in and after the 9th time, smoothing and all, the idle was still out 4% and only started about 6% out.

[Dirk Diggler]

http://69.251.81.142/Vetable.jpg
http://home.comcast.net/~lhas9978/log.jpg

[Black02SS]

I don't have my laptop with me or I would post a screen shot of mine as well. My VE table is smoother, but my ben factor doesn't look near as close. Now if I don't smooth the table, then it falls right into place. It appears the question again is, how smooth does this table have to be? We know the mafless GTS using a spikey table but we hear again how they interpolate and the table needs smooth...... What were you using here to smooth, 50%????

[Dirk Diggler]

I don't have my laptop with me or I would post a screen shot of mine as well. My VE table is smoother, but my ben factor doesn't look near as close. Now if I don't smooth the table, then it falls right into place. It appears the question again is, how smooth does this table have to be? We know the mafless GTS using a spikey table but we hear again how they interpolate and the table needs smooth...... What were you using here to smooth, 50%????

Yup 50.... 60 and 70 seemed to have me chasing the wind

[Black02SS]

I agree 100%. Got any ideas??

[Dirk Diggler]

LOL If i did I wouldnt be posting here and you would be the first to know.... :D

[Black02SS]

Ok your worthless, anyone else???

[Scoota]

Hey Guy's,
'Ok', it’s probably nice to see the VE's looking like a quilt blanket; it’s not going to be a point of embracement if you have a slight roughness in the 3D graph. We are tuning, "not displaying art work". The Cells will still interpolate well smoothing with multiple weights of about 10-20 at a time,
If you have used our mate "BEN" to do your Autotune, Then you sould trust his judgment without OVER smoothing.

Cheers Scotty.

[Black02SS]

So you are saying to only smooth by 10 to 20? Its early here. LOL

[GMPX]

The pic shows part throttle section of the VE table from my car, this runs 14.6 - 14.7 spot on in all these area's, it looks ugly, but thats what the engine and the WBO2 tell me what it needs.
I think smoothing the VE table might not be very wise if you have used the BEN factor, it is what it is!!.

However, I do think smoothing the spark maps is probably worth doing.

Cheers,
Ross

[Delco]

One thing to remember when using the ben factor is that it take the area around it data as well

lets take a 2000/50kpa point , it will use all the data from 1800-2200 and the data from the 57-62kpa region , if you only gather a small median of data the area can see a spike , also different air temps and atmospheric onditions will slew the data slightly. To see what I mean go and have a look at the min and max value for each point is.
You will also get transients creeping into the data.

The errors can be smoothed by taking a large chunk of data for each point. On my car whe have done approx 30 logs and used the data to get the table very smooth but each log has been about a hour+ of driving in lots off different conditions.


the ben factor is not a be all and end all to get the perfect tune , its to get you into the ballpark in a very small time frame , carefully smoothing ( not excessive ) and lots of data will get you closer to the right point.

Another way to look at this is to log the VE value , in hteory the map should come out exactly as it is in the calibration - it doesnt it has the same sort of spikes ( even though the calibration table may be completely smooth) due to the above mentioned problems.

Its the same with any auto tune feature - be it motec, autronic ,ame , kalmaker etc.

[SinisterSS]

My VE on the LS6 with headers, and 3" dual exhaust came out smoother than Ross's example. Consistent weather during AutoVE, careful throttle application, filtering of transients, and huge 1+ hour log files and several iterations got all my BENs within less than 1%. Took about 6 hours of logging and six iterations of the VE.

If there is a cell in the VE that comes up 120 surrounded by completely by cells reading 65, by all means, hand tweak the 120 cell and re-AutoVE and see where it comes up again.

The 120 example may be a real spike (usually not that big of a difference or limited to one cell) due to the harmonics of the intake, rapid throttle application, etc.

As you are SD AutoVE tuning and getting the BENs all close to 1.00, you will notice how nice the car runs - you can feel it - smooth power delivery from idle to redline at any manifold pressure.

Smoothing VE maps in the past has just had me chasing my tail.

[Dirk Diggler]

There is so much confilcitng information about a smooth ve table and a not smooth ve table. I think our frined "BEN" coupled with well chossen filter criteria will make up and get us damn close to a fairly smooth ve table....


Maybe tightening up the TP% change criteria will lead to a smaller more uniform BEN factors that wont creat that many spikes and valleys. 2% 2.5% maybe?


Also as far as smoothig is concened I think going from a stock table to tuning something like a full bolton and greater car. You can start out will 100% smoothing on the first pass to normalize everythign. Each addition pass on the ve table should decrease the smoothing weight. The closer you get the more you decrease the wieght. After a while it shouldnt really affectt the table anymore unless you are looking and the numbers to 6 decimal places

[SinisterSS]

Exactly. Smoothing works well when making that first AutoVE major tweak to the stock VE. Then just "let it ride" and see where it all shakes out.

A large portion of the tuning threads I see on LS1Tech are quite funny.

My favorite one of recent "What is the difference between open and closed loop?"

Put the laptop down, step away from the tuning software.

:lol:

[Black02SS]

GMPX/Blacky - If either of you get a chance, please read this thread..(starts really on page2)

http://www.ls1tech.com/forums/showthread.php?t=278819&page=1&pp=40

This is a conversation mainly between myself, DirkDiggler(HumpinSS), and Magnus regarding smoothing the ve table. Like Dirk mentioned above, why is their such a controversity over weather to smooth it or not? Keith says that it needs to be smooth because of the interpolation of the cells, where I have stated since day one like Ross did, its what the car wants and what is what the wideband tells me it needs. So is there really a correct way or is it just a personal preference? I have tried to smooth and make it look perty, but it just doesn't seem to work as my actual differes greatly from my commanded.

[GMPX]

My favorite one of recent "What is the difference between open and closed loop?"

Put the laptop down, step away from the tuning software.

:lol:

:lol: :lol: :lol: :lol:

[GMPX]

Keith says that it needs to be smooth because of the interpolation of the cells, where I have stated since day one like Ross did, its what the car wants and what is what the wideband tells me it needs. So is there really a correct way or is it just a personal preference? I have tried to smooth and make it look perty, but it just doesn't seem to work as my actual differes greatly from my commanded.

Keith has a point in that you would not get away with BIG differences between cells, smoothing is helpful there, but as Rick said (hope it was you Rick) your first log of the BEN factor might produce some spikey areas, smoothing on the first pass would be a good idea to get things in the ballpark, then logs done after that will be pretty much what the engine wants, it does depend on how crazy you want to get with this stuff, but because the PCM can interpolate (256 from memory) values between the cells so you are never going to hit the same points at say 2000 - 2400 and 55 - 60kPa, so between those cells the PCM may calculate a VE number of 1.203 or 1.209 or 1.198 etc (I did a thread on that somewhere), the more logging you do, the more correction you apply will pretty much reduce those error factors.
You are always fighting the PCM interpolating between the cells, the idea is to narrow the range of that interpolation by either 'measuring it' or 'smoothing' it which is a calculated correction.

I don't spend 8hrs a day tuning cars so I just can't say 100% that you should not use smoothing all the time, mathematically it doesn't make sense to me to do it that way. But practically I'd love to hear from others, I found on my own car making the graph look like a babies blanket screwed my WBO2 values.

Delco, if you wouldn't mind passing on your eperiences with your car that would be good, I know that thing has a BEN factor of what 1.0001 everywhere!!.

Cheers
Ross

[SinisterSS]

My BENs are around 1.00X in most cells with the worst one probably around 1.01. If I get a chance tonight, I'll post a screenshot of the VE. It is not perfectly smooth (never, ever smoothed it) and the car performs flawlessly and the LTFTs values are all exactly the same -1.51% by design in closed loop.

It takes literally hundreds of miles of driving, logging data with multiple VE iterations under controlled conditions and then using the filter to glean the data. My approach is to fill the entire BEN table with lots of counts in each cell and make a global update to the VE. Rinse and repeat until the BENs start coming up with values like 0.996, etc.

And no, I don't tune my MAF with LTFTs. :wink:

[Delco]

Keith says that it needs to be smooth because of the interpolation of the cells, where I have stated since day one like Ross did, its what the car wants and what is what the wideband tells me it needs. So is there really a correct way or is it just a personal preference? I have tried to smooth and make it look perty, but it just doesn't seem to work as my actual differes greatly from my commanded.

Keith has a point in that you would not get away with BIG differences between cells, smoothing is helpful there, but as Rick said (hope it was you Rick) your first log of the BEN factor might produce some spikey areas, smoothing on the first pass would be a good idea to get things in the ballpark, then logs done after that will be pretty much what the engine wants, it does depend on how crazy you want to get with this stuff, but because the PCM can interpolate (256 from memory) values between the cells so you are never going to hit the same points at say 2000 - 2400 and 55 - 60kPa, so between those cells the PCM may calculate a VE number of 1.203 or 1.209 or 1.198 etc (I did a thread on that somewhere), the more logging you do, the more correction you apply will pretty much reduce those error factors.
You are always fighting the PCM interpolating between the cells, the idea is to narrow the range of that interpolation by either 'measuring it' or 'smoothing' it which is a calculated correction.

I don't spend 8hrs a day tuning cars so I just can't say 100% that you should not use smoothing all the time, mathematically it doesn't make sense to me to do it that way. But practically I'd love to hear from others, I found on my own car making the graph look like a babies blanket screwed my WBO2 values.

Delco, if you wouldn't mind passing on your eperiences with your car that would be good, I know that thing has a BEN factor of what 1.0001 everywhere!!.

Cheers
Ross

SinisterSS is right on the money , lots of data , careful gleaning out of bad daa is what is required , one or two passes will only get you intot he ballpark.

One interesting thing I noted yesterday on my combo ( this is a development version I am using in full AlphaN in anticipation of big cams etc and 2 bar ) is that IAT can have a big effect , I did 2 1 hr plus drives yesterday to a job datalogging as always and the average IAT was in the low to mid 20 deg C , all my benfactors where within 0.9998 to 1.0002 but as soon as I got into the city and had a few stop starts the IAT climbed to 50 deg C and the tune went lean ( this is a full open loop tune at present) by about 1 AFR.

Transients also mess up the BenPLX factor big time , really need to filter the data a litttle more heavily to get more accurate results , also while at steady state on the dyno with a fixed base pulse width the observed AFR will jump around by about 0.4 AFR due to the combustion process so this is not all a exact science.

use the BenPLX with careful data interpritaion , a little mild smoothing and lots and lots of data at constant temps. It is not there to take the tune out of tuning but to help get to the base tune in a quicker time frame.

the other thing is you wont have a smooth VE table in the real world , it will have little ripples in it due to the inlet harmonics - for this reason I tend to use localised smoothing at a low factor rather than a blanket smooth.

[SinisterSS]

Here's the VE for my LS6 with 1.75" Tri-Y headers, 3" dual exhaust with x-pipe, and a honking big K&N filter. Needless to say, the VE wasn't even close in stock form.

This VE was created by six to eight iterations (didn't keep track) and has less than 1% error in the BEN, beautiful LTFTs and drivability.

Why smooth it?

[Black02SS]

Ok guys, I thought I was going nuts trying to get this stuff smooth. I did a log today and decided not to smooth, what do you know, my BEN was pretty close after one pass. I also noticed what Delco was refering to, my IAT's jumped up to 180*F (no C*) here, and my AFR went to the high 15's low 16's and I didn't know what had went wrong as the tune was the same, I just restarted my car. Could the timing reduction for IAT cause this or is it just the fact that the intake temps were so hot? I had my car dialed in when I was using HPT with a spikey table and was told that it was incorrect and needed to be blanket smooth which is why I tried re-doing it.

Could you guys specify more in depth on what you use to filter out the data? How many cell counts does one consider "engough" before you can move to the next cell? Obviously the more the better, but is there a magic number that works well for ya'll? I personally like to see ~50 counts in each cell for part throttle when doing the tune.

Thanks for all the replies and help.

[SinisterSS]

50 to 5000 counts. :D

Filter at different percentages, if you have 5000, start with a super fine one and see how many it counts it gets rid of. If you have 50, a more course filter.

[mistermike]

Smoothing? Not a problem.
http://www.cpmamerica.com/images/Asphalt06%20-%20Roller.jpg[/img]

[GMPX]

Nice burnout pad :lol: