Quick guide to simultaneous VVE/MAF calibration for gen4+

[GHuggins]

The formulas that HPT allows us to have are 'contextless'. What I mean by that, is they do not let us deal with (unless they've changed things and I missed them) data from any time other than right now. That's why we're limited to 'steady state'. In steady state, things don't change much, thus data from two samples ago is gonna be about the same as the data in the current calculation. This is why things like calibrating the MAT estimator's BIAS table is impossible inside of HPT. This has a further reaching consequences.

Here's a small snippet from what I wrote up back in 2008:
tuning the VE (or GMVE) consists of solving for GMVE:

GMVE= IFR*IPW*AFRwb*(IAT+(ECT-IAT)*BIAS)/MAP

This doesn't look too bad, right? Well, the kicker is that GMVE will be determined by BIAS, which by itself is determined by Airflow, which is determined by GMVE. You ever seen a dog chasing its tail?

You cant solve it separately, you MUST solve it simultaneously. But to do that, you must step away from HPT, it's just not doable.

Thanks Marcin for all your early and current work,

This is why I came up with the simple formula for just "making" the VE the same as MAF with fuel corrections considered into it. Actually works really really good. The longer VE formula's using fuel or airmass would always get different results log to log and I didn't like that "chasing you tail" feeling. I posted the formula above - it's a simple math just to make VE and MAF the same with fuel corrections figured into the MAF portion

[Alvin]

consists of solving for GMVE:

GMVE= IFR*IPW*AFRwb*(IAT+(ECT-IAT)*BIAS)/MAP

This doesn't look too bad, right? Well, the kicker is that GMVE will be determined by BIAS, which by itself is determined by Airflow, which is determined by GMVE. You ever seen a dog chasing its tail?

You cant solve it separately, you MUST solve it simultaneously. But to do that, you must step away from HPT, it's just not doable.

Current thinking is IAT ECT and bias /with filter is loggable on gen 4+ with manifold temp pid. Which does seem correct. Manifold temp on GEN 3's is not loggable (at this time, maybe never)

I've said it a bunch all over this forum. Calculating GMVE from MAF with or without fuel trim or Eq error compensation still outputs questionable results. IMHO MAF is just too erratic. Sometimes works decently, sometimes not.


I did a GEN 5 not too long ago that literally dug itself into a hole trying to do GMVE from MAF. Did one a week later the old fashion way by failing the MAF but then checked results afterwords and it looked pretty close.

[smokeshow]

Thanks Marcin for all your early and current work,

This is why I came up with the simple formula for just "making" the VE the same as MAF with fuel corrections considered into it. Actually works really really good. The longer VE formula's using fuel or airmass would always get different results log to log and I didn't like that "chasing you tail" feeling. I posted the formula above - it's a simple math just to make VE and MAF the same with fuel corrections figured into the MAF portion

Feel like you'd have issues with using strictly the reported VE airflows. The 512g/s VE airflow cap in the scanner and losing VE altogether in high speed mode are what immediately come to mind. I guess, unless you never rev high enough for high speed mode and are always under 512g/s lol. The log snip in the picture shows a common scenario where those parameters would not be useful.

Capture.PNG

[smokeshow]

Current thinking is IAT ECT and bias /with filter is loggable on gen 4+ with manifold temp pid. Which does seem correct. Manifold temp on GEN 3's is not loggable (at this time, maybe never)

I've said it a bunch all over this forum. Calculating GMVE from MAF with or without fuel trim or Eq error compensation still outputs questionable results. IMHO MAF is just too erratic. Sometimes works decently, sometimes not.


I did a GEN 5 not too long ago that literally dug itself into a hole trying to do GMVE from MAF. Did one a week later the old fashion way by failing the MAF but then checked results afterwords and it looked pretty close.

That sounds like it may be a symptom of MAF reversion. I mentioned it in the post here... https://forum.hptuners.com/showthrea...l=1#post657362

Don't take the log data as unconditional truth. The more wild your modifications, the less capable the control method is going to be for your setup. Key is just knowing how to spot it. If your MAF signal looks like a fuzzy caterpillar, probably a good sign you may not be able to use this method. And sometimes, you can't use that MAF config for fueling at all. I just had to have a guy swap his intake out because the MAF signal was garbage and no amount of trickery in the toon would be able to successfully get around it. Sometimes the hardware doesn't play ball.

[redhardsupra]

The moment anyone says 'sometimes it works' what they really mean is 'my tuning tools do not allow me remove the noise, or isolate the good parts to work with.' never stepping outside of HPT limits you greatly. You can't do good filtering and smoothing. you cant do robust statistics. you cant ask for a median instead of a mean. The list is long.

THE reason why I wrote up the 'three airmass models' as one thing, not as three separate blurbs, was so you can see how to mix and match them, depending on what your situation might be. Use the two that work great to create the third one. Or calibrate them separately, and then compare them individually; since they're all estimating the same entity, they should yield very similar results. I've shown how to do that in some other posts. This was meant as a toolkit, not a cookbook to follow.

[GHuggins]

Feel like you'd have issues with using strictly the reported VE airflows. The 512g/s VE airflow cap in the scanner and losing VE altogether in high speed mode are what immediately come to mind. I guess, unless you never rev high enough for high speed mode and are always under 512g/s lol. The log snip in the picture shows a common scenario where those parameters would not be useful.

Capture.PNG

I haven't run into a problem with it yet, but I'll keep an eye out. Also haven't used it for any wot corrections so probably the whole reason why it hasn't shown it's head. I've just been using it against fuel trim corrections, so the second it's going into PE and hitting those higher airflow modes it's not used. I'm using the airmass calculator from that point.

[Alvin]

The moment anyone says 'sometimes it works' what they really mean is 'my tuning tools do not allow me remove the noise, or isolate the good parts to work with.' never stepping outside of HPT limits you greatly. You can't do good filtering and smoothing. you cant do robust statistics. you cant ask for a median instead of a mean. The list is long.

THE reason why I wrote up the 'three airmass models' as one thing, not as three separate blurbs, was so you can see how to mix and match them, depending on what your situation might be. Use the two that work great to create the third one. Or calibrate them separately, and then compare them individually; since they're all estimating the same entity, they should yield very similar results. I've shown how to do that in some other posts. This was meant as a toolkit, not a cookbook to follow.

Do you have tools to remove the noise from the stock market too?

My point is often the MAF is too noisy for any sort of decent result backing out a VE table from MAF. It's far better to calibrate them independently.

[smokeshow]

Do you have tools to remove the noise from the stock market too?

My point is often the MAF is too noisy for any sort of decent result backing out a VE table from MAF. It's far better to calibrate them independently.

Noisy signals are still an issue if you calibrate them separately. So that's more a problem with MAF placement and intake design (and depending on the application, the Kalman filter tuning) than the choice in method for calibrating an airflow model. There are a number of ways to go about it, but all of them require trustworthy sensors and measurements.

[redhardsupra]

It's far better to calibrate them independently.

I agree. When two independent methods arrive at the same solution that gives us a lot of assurance.

But when MAF signal is too noisy, you can map SD-derived airflow values onto the MAF calibration. And the two independent calibrations are a perfect source of data to see under what set of conditions MAF is starting to produce nonsensical numbers.

It's a toolkit, not a cookbook.

[Alvin]

I think many get too lost in the hypothetical and theory instead of what is practical/effective in real world.

To know that both airflow models should end up the same BUT at the same time understand why little trust should be placed in one because of the limitations of how it works is key.


To speak as if a filter can be written to be used from one car to the next to address how this person has a sharp bend before his MAF, or this person has a large cam and the reversion is causing the MAF to read higher than it should, or this guys filter is behind the electric fan and starts to go nuts anytime the fan is on. blow off valves, and carbon fouling, reversion, air blown over vehicle in motion... all of it matters.

[smokeshow]

I think many get too lost in the hypothetical and theory instead of what is practical/effective in real world.

To know that both airflow models should end up the same BUT at the same time understand why little trust should be placed in one because of the limitations of how it works is key.


To speak as if a filter can be written to be used from one car to the next to address how this person has a sharp bend before his MAF, or this person has a large cam and the reversion is causing the MAF to read higher than it should, or this guys filter is behind the electric fan and starts to go nuts anytime the fan is on. blow off valves, and carbon fouling, reversion, air blown over vehicle in motion... all of it matters.

Sounds like you guys are in aggressive agreement with each other lol. I don't think anyone has claimed that there is a 'magic bullet filter' or anything though. And certainly it is not the intent of this simultaneous procedure. It just provides a tool to exploit the overlap between the MAF and the speed density model when when they both align with reality. But when you can't trust the MAF....don't. Just know that when the MAF gives you issues, its not just providing bad data for speed density correction...its like that all the time.

[GHuggins]

For what it's worth, I had one that I forgot to change the dynamic settings on and as a result the VE was 10% overly rich in the idle regions and then shifted 10% overly lean by 2000 rpms. Changing the dynamic settings back to "theoretical" 100% MAF got me better data acquisition while I re-corrected the MAF curve. Just saying the way it's setup can be just as much at fault as anything. If the MAF has that much "noise" on a particular build then it should probably either be corrected or made a SD tune regardless.

[smokeshow]

Yeah the prediction can absolutely work against you after a few engine mods. Late gen4 and on have MAF reversion rejection built right into those coefficients. Doesn't take much for them to be completely wrong, even with something as mild as a cold air intake.

I think the real solution here is to tease Marcin with the juicy math problem that he knows this is and talk him into making a nice little app that can tune the kalman filter. I can help with the algo...but I'm not writing any python or VBA. Lol

[ttz06vette]

Yeah the prediction can absolutely work against you after a few engine mods. Late gen4 and on have MAF reversion rejection built right into those coefficients. Doesn't take much for them to be completely wrong, even with something as mild as a cold air intake.

I think the real solution here is to tease Marcin with the juicy math problem that he knows this is and talk him into making a nice little app that can tune the kalman filter. I can help with the algo...but I'm not writing any python or VBA. Lol

Would love to see that!

[ttz06vette]

Way back when Abaco made a tunable MAF that allowed you to custom configure your MAF output curve and suppress outlier noise caused by intake bends or other airflow issues. They have since gone out of business but I still use their MAF. Options for tuning have improved so much since 2007 for Forced Induction and such but it still works.

[redhardsupra]

I think the real solution here is to tease Marcin with the juicy math problem that he knows this is and talk him into making a nice little app that can tune the kalman filter. I can help with the algo...but I'm not writing any python or VBA. Lol

:glare: Just when I thought I was out, they pull me back in.


Get me some data, and I'll let's take this out of HPT's incompetent hands, and into the 21st century.

[Cringer]

I have used this method with pretty decent success. However, I eventually ran into a brick wall with making progress. I am not sure if it is the MAF reversion issue mentioned above (I do have a healthy cam), or something else. What I saw was that I could never really get the MAF totally dialed in as well I thought I could and when looking at the GMVE_CL plotted on a histogram, I was getting really spiky lean spots every time at 3,000 - 3,200 RPMs, even though fuel trims were still inline and not reporting lean. As a test I tried to increasing the VVE that area many times (which resulted in poor transition), however, the lean spike persisted. So I failed the MAF and went traditional SD and found that the lean spot was gone/did not exist.

From reading above it seems like this method can be different depending on the ECM, engine build, and MAF location/intake tubing...and that may be my issue...I can't imagine a whole lot (more) reversion happening at 3k RPMs though. But in general, since this is a calculated value from many sensors and other calculated values (perhaps the Manifold Air Temp is the culprit), the value seems to be all over the place.

[ttz06vette]

So would this tool work with a 2bar OS upgrade and VE vs. VVE if you flipped the axis’s and input the zones?

[cpd004]

Interested to see this!

[dubbledtadab]

1 way to calculate GMVE.

Here is how I did it a while back. Without MAT I found the bias of 0.35 worked for me. The bias also changes depending on how your driving. E.g, for power pulls, your bias could start out at 0.35 but by the end of the pull be down at 0.31. Spent way to many hours on this. Trust me here, best is with MAT.

You need 2 formulas. One for closed loop, and another for open loop where your commanded AFR is used.

This formula sitting inside smokeshows formula ends up been pretty nasty to look at. Especially worse if you make an error, then trying to work out where you mucked up

Without MAT
GMVE = IFR * IPW * AFR * (IAT + (ECT - IAT) * BIAS) / MAP

With MAT
GMVE = IFR * IPW * AFR * MAT / MAP

IFR g/sec Injector flow rate
IPW msec Injector pulse width
MAT degK Manifold air temperature
IAT degK Intake air temperature
ECT degK Engine coolant temperature
MAP kPa Manifold air pressure
BIAS Constant, 0.2875 -> 0.35

Required channels:
Manifold Air Temp
Engine Coolant Temp (if no manifold air temp available)
Intake Air Temp (if no manifold air temp available)
Intake Manifold Absolute Pressure
Injector Pulse Width Avg.
Injector Flow Rate

Credit to: http://redhardsupra.blogspot.com/200...odels.html?m=1

What are the formulas for open and closed loop?