How to Scale your MAF for Flash Tuning (Cobb, Hymee)
#151
Boosted Kiwi
iTrader: (2)
Just fixed most of this since Kane told me about the Ve chart . It is close enough to spot on now for me to be a happy chappy .
My maf curve looks fn sweet as well now as I've finally cracked all the Greddy wastegate/boost control issues . Not to mention that I'm certainly pushing well over 320whp now
Last edited by Brettus; 07-05-2009 at 09:50 PM.
#152
U-Stink-But-I-♥-U
iTrader: (1)
Join Date: Mar 2005
Location: 12 o'clock on the Beltway.
Posts: 2,004
Likes: 0
Received 1 Like
on
1 Post
Just fixed most of this since Kane told me about the Ve chart . It is close enough to spot on now for me to be a happy chappy .
My maf curve looks fn sweet as well now as I've finally cracked all the Greddy wastegate/boost control issues . Not to mention that I'm certainly pushing well over 320whp now
My maf curve looks fn sweet as well now as I've finally cracked all the Greddy wastegate/boost control issues . Not to mention that I'm certainly pushing well over 320whp now
#153
rot8ing
iTrader: (1)
carbon, i doubt anyone has thought as hard about tuning as you. i appreciate your efforts.
interesting stuff especially with respect to the lorenz curve.
as to your second graph, I'll think more about it tonight, but best I can tell you're making sense. i would go ahead and run the experiment since I'm not sure what else there is to account for.
interesting stuff especially with respect to the lorenz curve.
as to your second graph, I'll think more about it tonight, but best I can tell you're making sense. i would go ahead and run the experiment since I'm not sure what else there is to account for.
#154
U-Stink-But-I-♥-U
iTrader: (1)
Join Date: Mar 2005
Location: 12 o'clock on the Beltway.
Posts: 2,004
Likes: 0
Received 1 Like
on
1 Post
That is very kind but absolutely untrue. I am a rank amateur at best. Thank you, though. Gave me warm fuzzys.
I dont know what you meant about "interesting" with regards to the lorenz curve. If you are following the tacit implications of what I was suggesting, we should talk.
I will try the modification to the injector scalars later this week and see what happens. Thanks for the encouragement!!
I dont know what you meant about "interesting" with regards to the lorenz curve. If you are following the tacit implications of what I was suggesting, we should talk.
I will try the modification to the injector scalars later this week and see what happens. Thanks for the encouragement!!
#155
Boosted Kiwi
iTrader: (2)
Carbon ,
What i've done is basically utilise "boost creep" by restricting how far the wastegate can open - it gives me progressively increasing boost , unlike the stock Greddy which is the exact opposite . The g/s curve is now a thing of beauty and it drives so much better at part throttle compared to when i was using a boost controller. I still use the secret squirell controller BTW except now it operates the boost controller from 6500 onwards so i can dial it up to the max in that range .
Boost controllers suck at controlling this turbo
The Ve table i had was probably very different to what comes with the AP (looking at what Kane posted) so it may not be an issue for you . The issue was always WOT vs part throttle - cruise was never an issue.
I basically adjusted the ve table to correct target vs actual in the areas i have been fighting with
What i've done is basically utilise "boost creep" by restricting how far the wastegate can open - it gives me progressively increasing boost , unlike the stock Greddy which is the exact opposite . The g/s curve is now a thing of beauty and it drives so much better at part throttle compared to when i was using a boost controller. I still use the secret squirell controller BTW except now it operates the boost controller from 6500 onwards so i can dial it up to the max in that range .
Boost controllers suck at controlling this turbo
The Ve table i had was probably very different to what comes with the AP (looking at what Kane posted) so it may not be an issue for you . The issue was always WOT vs part throttle - cruise was never an issue.
I basically adjusted the ve table to correct target vs actual in the areas i have been fighting with
Last edited by Brettus; 07-05-2009 at 11:29 PM.
#157
Sure you can log your airflow. But you are just logging what the MAF calibration curve says. The ECU only sees the voltage of the MAF. From that, it looks at the MAF calibration and deduces what the airflow is. The only way you can do it is to use a know calibrated mass flow meter and compare to the sample. You could run 2 x MAFS, one connected to the ECU, and one just logging voltage independantly, but it would all be relative to the accuracy of the first MAF.
Cheers,
Hymee.
Cheers,
Hymee.
One other thought. Carbon, as far as the injectors, are you only playing with injector scaling, latency, or both? I would assume that published latencies for the injectors are pretty much set in stone. You could use the same scaling:injector size ratio as stock for the bigger injectors. Again, I'm just trying to eliminate one of your tuning variables (injectors, MAF, VE). Injector scaling should cause global AFR changes. But it sounds like you are dealing with "bumps" in your tuning curves. So MAF and VE changes would be more appropriate.
I tune Evos and WRX/STis so please excuse me if I am making incorrect RX8 tuning assumptions. I'm friends with Carbon and I am just trying to help out.
#158
U-Stink-But-I-♥-U
iTrader: (1)
Join Date: Mar 2005
Location: 12 o'clock on the Beltway.
Posts: 2,004
Likes: 0
Received 1 Like
on
1 Post
Yeah, the latency tables are done. I based them on Injector-rehab.com data. To get any more info on those, I would have to send them out to someone.
The scaling is based on the ratio of the ATR number for the stock injector size to the larger injector size. That assumes that the scaling factor is linear and that all the flows are known accurately. My guess is that the expected flows vs the actual flows are different, and I get to guess what the actual flows actually are.
My bumpy maf curve does a good job at making the best of things and the car is running well, but I figure as I am going to start leaning things up and uping the boost soon, I should get the base tune as good as it is going to get.
BTW, I tested my vac at idle and at 2krpm and got around 18"Hg and 20"Hg, respectively. I will try to get my compression numbers this week so that I can watch them as I go to higher loads. Hoping to get 10-20k miles more out of this motor. I appretiate the input.
The scaling is based on the ratio of the ATR number for the stock injector size to the larger injector size. That assumes that the scaling factor is linear and that all the flows are known accurately. My guess is that the expected flows vs the actual flows are different, and I get to guess what the actual flows actually are.
My bumpy maf curve does a good job at making the best of things and the car is running well, but I figure as I am going to start leaning things up and uping the boost soon, I should get the base tune as good as it is going to get.
BTW, I tested my vac at idle and at 2krpm and got around 18"Hg and 20"Hg, respectively. I will try to get my compression numbers this week so that I can watch them as I go to higher loads. Hoping to get 10-20k miles more out of this motor. I appretiate the input.
#159
Hey Carbon;
Yes I had reviewed other than normal distributions.... however; as we approach N then we will almost always approach a normal distribution with almost any data set (law of large numbers). So I exclude outliers from the AFR logged areas not the MAF scale itself.
PS - Are you accounting for your sensor latency in your calculations? It makes quite the difference in transitional areas of the map.
Again MAF scale issues show up as deviation around a specific voltage - injector scale issues should be shoing up as a specific variance at load and RPM; regardless of MAF voltage. And anything else that cannot be resolved would go over to the VE table - IMO.
Your graphs are cool - much sexier than mine - LOLZ
Yes I had reviewed other than normal distributions.... however; as we approach N then we will almost always approach a normal distribution with almost any data set (law of large numbers). So I exclude outliers from the AFR logged areas not the MAF scale itself.
PS - Are you accounting for your sensor latency in your calculations? It makes quite the difference in transitional areas of the map.
Again MAF scale issues show up as deviation around a specific voltage - injector scale issues should be shoing up as a specific variance at load and RPM; regardless of MAF voltage. And anything else that cannot be resolved would go over to the VE table - IMO.
Your graphs are cool - much sexier than mine - LOLZ
#160
U-Stink-But-I-♥-U
iTrader: (1)
Join Date: Mar 2005
Location: 12 o'clock on the Beltway.
Posts: 2,004
Likes: 0
Received 1 Like
on
1 Post
The Lorentz or Cauchey distribution is another thing altogether. This population has as a fundamental characteristic of a propensity for members of the population that posess large differences from a median or mode. This distribution has very large wings, or bias to both sides of a central value. The deeper result of this is that a mean or average is not defined. THERE IS NO MEAN! You can calculate it, but it doesn't mean (no pun intended) what mean means (ok, I intended it that time.)
Now, how this affects us, if at all, may be minor. It also may be that what I found is anomalous or just plain poorly analyzed. In any case, I would think that a bimodal distribution would confuse your software far more than a Lorentz, but what concerns me is that there is a small possibility that an analysis that uses variance to reach a conclusion from such a data set may reach an erroneous one. A simple visual test would catch anything, of course.
Nevertheless, I find this very interesting and am looking into what it means, if anything.
#161
Once again you are correct - however....
Given the noise in the sensors is random (MAF, O2, Injector PW etc); otherwise you would always hit the target AFR's right?
So the law of large numbers does apply to random values....
Either way - I think we are talking semantics unless you want to change my software distribution curves and test.... cause I am inside 1% on most cars... I'll live with it.
Given the noise in the sensors is random (MAF, O2, Injector PW etc); otherwise you would always hit the target AFR's right?
So the law of large numbers does apply to random values....
Either way - I think we are talking semantics unless you want to change my software distribution curves and test.... cause I am inside 1% on most cars... I'll live with it.
#162
U-Stink-But-I-♥-U
iTrader: (1)
Join Date: Mar 2005
Location: 12 o'clock on the Beltway.
Posts: 2,004
Likes: 0
Received 1 Like
on
1 Post
Don't get me wrong, I am not dinging you work in any way, shape or form. I think what you have created is an amazing bit of software. I am also not trying to change how you or anyone tunes. I am not even strongly suggesting that I even have anything useful to contribute, particularly with respect to this whole distribution/statistics thing.
I am just curious to understand what is contributing to the observed distribution, if it exists. So as you suggested, one possible sourse of error is the latency, or just goddam random error. All important is certainly in existance.
However, what I suspect is that the observed distribution, with a rich smattering of "outliers" is a result of interaction that betrays the existance of feedback or feedforward algorithms in openloop that we have yet to understand. And if it is not this, then some other physical interaction that might be interesting, again if the observation holds. If so, there might be something to learn and better statistical methods to apply.
I dont expect anyone to be interested or expect that any of this is critical. Hell, most people tune on a wing and a prayer. If it works, fine. I would suppose that such folk would suggest that I am putting too much thought into this whole thing. I suppose I am. Whatever.
So yeah. What you are doing works. I dont disagree a bit.
I am just curious to understand what is contributing to the observed distribution, if it exists. So as you suggested, one possible sourse of error is the latency, or just goddam random error. All important is certainly in existance.
However, what I suspect is that the observed distribution, with a rich smattering of "outliers" is a result of interaction that betrays the existance of feedback or feedforward algorithms in openloop that we have yet to understand. And if it is not this, then some other physical interaction that might be interesting, again if the observation holds. If so, there might be something to learn and better statistical methods to apply.
I dont expect anyone to be interested or expect that any of this is critical. Hell, most people tune on a wing and a prayer. If it works, fine. I would suppose that such folk would suggest that I am putting too much thought into this whole thing. I suppose I am. Whatever.
So yeah. What you are doing works. I dont disagree a bit.
#164
I wasn't knocking you at all Carbon - just going with my theories and findings.
Question for you - any chance you have enough data (+100K) records on the same tune that I could put into baseline and see if it is random or if there is a set trend one way or the other? I am down to try anything.
Question for you - any chance you have enough data (+100K) records on the same tune that I could put into baseline and see if it is random or if there is a set trend one way or the other? I am down to try anything.
#165
U-Stink-But-I-♥-U
iTrader: (1)
Join Date: Mar 2005
Location: 12 o'clock on the Beltway.
Posts: 2,004
Likes: 0
Received 1 Like
on
1 Post
Most of the time i gather about 10k filtered data points per tune (over, perhaps, 3 drive cycles; I might start with 100k of CL and OL points but it usually gets whittled down after selecting OL only, above 3000rpm and load>0.2) before I deside to change something. So, not yet. After this next iteration, we can get together and check something out. I would be happy to collaborate.
Right now, I am supposed to be writing a grant application, so I am going to be a little slow.
Brettus, I appretiate the sentiment.
Right now, I am supposed to be writing a grant application, so I am going to be a little slow.
Brettus, I appretiate the sentiment.
#169
Lambda Can be used easily.
Latency that I have seen with flashers is somewhere areoung 500ms...but you should test it to be sure.
Commanded EQ is your target lambda and the O2B is your actual. I am not sure how much to trust the commanded eq to be acurate cause I've never used it.
Latency that I have seen with flashers is somewhere areoung 500ms...but you should test it to be sure.
Commanded EQ is your target lambda and the O2B is your actual. I am not sure how much to trust the commanded eq to be acurate cause I've never used it.
#170
Ultra Noob
iTrader: (1)
Ok, I have a stupid question for Kane. In your spreadsheet, you calculate the percent difference from lamda. Then it appears that you subtract the percent difference from the g/s in order to get the actual g/s. Can you please explain to me the relation between the two? Wouldn't you want to multiply them by the percent difference then subtract/add based on what your result would be? I'm sorry if this has been asked before, I just don't quite understand and don't want muck up my engine because of it. Thanks
#171
Ok, I have a stupid question for Kane. In your spreadsheet, you calculate the percent difference from lamda. Then it appears that you subtract the percent difference from the g/s in order to get the actual g/s. Can you please explain to me the relation between the two? Wouldn't you want to multiply them by the percent difference then subtract/add based on what your result would be? I'm sorry if this has been asked before, I just don't quite understand and don't want muck up my engine because of it. Thanks
So the percentage adjustment made in relation to afr = percentage change in airflow.
air/fuel = gsec/inj duration
#172
Ultra Noob
iTrader: (1)
OK, That certainly helps. But I'm still not sure why the actual g/sec is simply the g/sec - %dif of lambda. I thought it would be a percentage as well. For example, using the first line I thought it would be 5g/sec*.98%=4.90actual g/sec
Upon re-reading your comment. It seems like we both mean the same thing., but the numbers in your spreadsheet don't match the math.
Upon re-reading your comment. It seems like we both mean the same thing., but the numbers in your spreadsheet don't match the math.
Last edited by bwilk; 11-05-2012 at 04:38 PM. Reason: Phones suck at forums
#173
OK, That certainly helps. But I'm still not sure why the actual g/sec is simply the g/sec - %dif of lambda. I thought it would be a percentage as well. For example, using the first line I thought it would be 5g/sec*.98%=4.90actual g/sec
Upon re-reading your comment. It seems like we both mean the same thing., but the numbers in your spreadsheet don't match the math.
Upon re-reading your comment. It seems like we both mean the same thing., but the numbers in your spreadsheet don't match the math.