Tuning MAF and VVE at the Same Time

[smokeshow]

Looking briefly at your log, all of your airflow data is captured at a 250ms raster, so you're likely missing the transient that would otherwise be glaringly obvious. Speed up that sample frequency and it'll become clear why dynamic airflow shouldn't be used for fueling error calculation. I should note that it can't be MAF only to see the discrepancy.

This is a throttle tip-in on a blower car I worked on recently. Airflow here is captured at 50ms sample time. As soon as TPS activity violates the SS criteria, airflow follows the VE airflow. When TPS stabilizes, it blends from VE airflow to MAF. As you can tell, dynamic airflow matches neither MAF or VE airflow when it is blended...so fueling error in between those points will be inaccurate for all steady state models. Again, it goes back to the inability to actually measure the airflow in the transient...so it has to be estimated. For scenarios (and applications) where the MAP and TPS prediction control is active (the Kalman filter), this discrepancy between the estimation and modeled airflow gets even higher.

Untitled.png

[Cringer]

I am going to disagree with you. Here a screenshot from the log I posted above. You can clearly see throttle is at steady state (29%), and Dynamic Air is matching VVE NOT MAF. And the MAF is not in a failed state.

You can also see at this frequency MAF is really lean as I intentionally did for this test.
And you can see the fuel trims are -19%...which is where I made the VVE table intentionally 20% rich.
All this is to say, Dynamic Air is the final boss. It only takes input from MAF and VVE + Transient regardless of steady state or transient throttle.
Maybe this is unique to my E38 and all other ECM's are different?

To be honest, I am not worried about the transient stuff. If you use dynamic air the way I am proposing you can use it as the benchmark to dial in the MAF and VVE at the same time and do it accurately and in less logging sessions/corrections.

Look at this screenshot and think about this. What will happen when a tuner sees that his fuel trims are 19% rich? He is going to remove 19% from the MAF...but holy shit batman...the MAF IS LEAN TO BEGIN WITH and will detune the system even more after they copy/paste special! All my method is doing is saying, if Dynamic air is 49 g/s and fuel trims 19% rich, then both MAF and VVE should be 39.69 g/s at this point...so update each accordingly. This system is working beautifully for me.


a5YUUDN.jpg

[Cringer]

Looking briefly at your log, all of your airflow data is captured at a 250ms raster, so you're likely missing the transient that would otherwise be glaringly obvious. Speed up that sample frequency and it'll become clear why dynamic airflow shouldn't be used for fueling error calculation. I should note that it can't be MAF only to see the discrepancy.

This is a throttle tip-in on a blower car I worked on recently. Airflow here is captured at 50ms sample time. As soon as TPS activity violates the SS criteria, airflow follows the VE airflow. When TPS stabilizes, it blends from VE airflow to MAF. As you can tell, dynamic airflow matches neither MAF or VE airflow when it is blended...so fueling error in between those points will be inaccurate for all steady state models. Again, it goes back to the inability to actually measure the airflow in the transient...so it has to be estimated. For scenarios (and applications) where the MAP and TPS prediction control is active (the Kalman filter), this discrepancy between the estimation and modeled airflow gets even higher.

Untitled.png

Again I will repeat.

It does not matter if Dynamic Air is 100% accurate all all times. Nor does it matter if it is mirroring VVE, or MAF, or if we are steady state, or in transient, or somewhere in-between.

The only thing that matters is that the fuel trims are responding only to Dynamic Air, and only Dynamic Air.

Thus we cannot apply fuel trims directly back to VVE and MAF since there is an indirect relationship between these two airflow models and Dynamic Air.

However, if we know the error correction (fuel trims), then we can apply it to the known starting point (Dynamic Air).

Then we can use that result as the standard to apply back to MAF and VVE in a true simultaneous fashion.

[smokeshow]

However, if we know the error correction (fuel trims), then we can apply it to the known starting point (Dynamic Air).

You're gonna have to explain that a bit more. How would you even know what fuel trim data is relevant if you're just capturing any and all error willy nilly lol. You sound confident, but you're missing the idea here...if you don't know where the fueling error originated from, there is no 'apply mack to MAF and vve'.

[GHuggins]

I've got a question and mean absolutely nothing ill by it - just a question. It's obvious that there is no 100% way to take VE out of the MAF fueling - you can have MAF setup to be the theoretical 100% via dynamic settings and dial it in then dial in your VVE then you'll find the MAF is again off even though it shouldn't be as nothing other than VVE changed. Is there a true 100% way to make it MAF only - 0ing out tables, setting certain tables to 1's or whatever? I make changes to VVE using VE error to MAF and after making a few passes I'm able to get MAF and VVE closer to one another and I take it that this is what Cringer is doing, but VVE won't be perfect for fueling error this way. I understand pedal movement affects it as transients come into play at that point, but is there a real true way to be 100% MAF without other factors playing into it or are filters the only answer?

[Cringer]

You're gonna have to explain that a bit more. How would you even know what fuel trim data is relevant if you're just capturing any and all error willy nilly lol. You sound confident, but you're missing the idea here...if you don't know where the fueling error originated from, there is no 'apply mack to MAF and vve'.

I can't believe we are going in circles like this. I have posted my issues with using the GMVE formulas, admitting I may have set them up wrong. Fortunately, Capt Chewy came out of the woodwork and PM'd me his formula. It definitely was an improvement, yet still resulted in a strange lean spot around 3000-3200 RPMs. When I fail the MAF and go into SD there is no lean spot. Hmm something is funky with the GMVE formulas.

While I do not expect you to examine my tune and logs and "fix them up" using GMVE theory, you certainly have had all that is needed to do that and change my mind. But instead, you just keep posting over and over "it doesn't work that way" without providing any real tangible info other than "trust me bro".

I have posted my tunes and logs. I have pointed out where Dynamic Air is in fact using VVE instead of MAF at steady state. No response from you on that either.

At the end of the day I have applied the scientific method here testing multiple theories and examined the results. From my point of view Dynamic Air is the clear winner. And why am I so confident? Because the ultimate test of the results are looking at the actual fuel trims themselves...NOT looking at the results filtered through the lens GMVE formula or the Dynamic Air formula, but just pure simple unadulterated STFT + LTFT data. It is all about the actual fuel trims period. My question is, why are you so confident about GMVE? I tested it and maybe it works in some cases, but it certainly finicky with my car. And if GMVE is sensitive to large cams, MAF reversion, etc...then perhaps it is not the ideal model to be using.

I mean would it be so difficult for you to create a formula or two and run them along side each other in the scanner? WTF. It is free. You could so easily test both methods side by side.

[Cringer]

I've got a question and mean absolutely nothing ill by it - just a question. It's obvious that there is no 100% way to take VE out of the MAF fueling - you can have MAF setup to be the theoretical 100% via dynamic settings and dial it in then dial in your VVE then you'll find the MAF is again off even though it shouldn't be as nothing other than VVE changed. Is there a true 100% way to make it MAF only - 0ing out tables, setting certain tables to 1's or whatever? I make changes to VVE using VE error to MAF and after making a few passes I'm able to get MAF and VVE closer to one another and I take it that this is what Cringer is doing, but VVE won't be perfect for fueling error this way. I understand pedal movement affects it as transients come into play at that point, but is there a real true way to be 100% MAF without other factors playing into it or are filters the only answer?

I wonder what effect unplugging the MAP sensor would have? Would it impact other areas than just the VVE table? Would that force it to go into MAF only?

I fear zeroing out the VVE tables is going to really throw the airflow model out of whack.

there is no 100% way to take VE out of the MAF fueling

This is my point...this is focusing on the wrong thing. Focus on Dynamic Air and fuel trims and then apply that data back to VVE and MAF. There no need to try to separate MAF and VVE.

[smokeshow]

I am going to disagree with you. Here a screenshot from the log I posted above. You can clearly see throttle is at steady state (29%), and Dynamic Air is matching VVE NOT MAF. And the MAF is not in a failed state.

You can also see at this frequency MAF is really lean as I intentionally did for this test.
And you can see the fuel trims are -19%...which is where I made the VVE table intentionally 20% rich.
All this is to say, Dynamic Air is the final boss. It only takes input from MAF and VVE + Transient regardless of steady state or transient throttle.
Maybe this is unique to my E38 and all other ECM's are different?

To be honest, I am not worried about the transient stuff. If you use dynamic air the way I am proposing you can use it as the benchmark to dial in the MAF and VVE at the same time and do it accurately and in less logging sessions/corrections.

Look at this screenshot and think about this. What will happen when a tuner sees that his fuel trims are 19% rich? He is going to remove 19% from the MAF...but holy shit batman...the MAF IS LEAN TO BEGIN WITH and will detune the system even more after they copy/paste special! All my method is doing is saying, if Dynamic air is 49 g/s and fuel trims 19% rich, then both MAF and VVE should be 39.69 g/s at this point...so update each accordingly. This system is working beautifully for me.


a5YUUDN.jpg

Are you going to tell the class about the DTC that showed up just before this screen shot? You know, the one that resulted in the injector pulse width doubling and flooding the engine with fuel as it failed the rationality diagnostic... I know you felt it. I can see it in the RPM and speedometer trace. Yeah, dynamic airflow is guaranteed to match the VE airflow when your experiment with the VVE and MAF cal forces it to ignore the MAF because it suspects hardware failure. I'm not surprised you never got back into the throttle after that little incident...puckered the asshole a little bit, eh?

Don't be dishonest dude. We can literally see it in the data.

[Cringer]

Are you going to tell the class about the DTC that showed up just before this screen shot? You know, the one that resulted in the injector pulse width doubling and flooding the engine with fuel as it failed the rationality diagnostic... I know you felt it. I can see it in the RPM and speedometer trace. Yeah, dynamic airflow is guaranteed to match the VE airflow when your experiment with the VVE and MAF cal forces it to ignore the MAF because it suspects hardware failure. I'm not surprised you never got back into the throttle after that little incident...puckered the asshole a little bit, eh?

Don't be dishonest dude. We can literally see it in the data.

Hell yeah I felt it and it scared the shit out of me for sure! But then again dude, this is a result of me intentionally jacking up the airflow model to teach the class! This does not happen when I roll around on my normal tune. That all you got?

[Cringer]

Are you going to tell the class about the DTC that showed up just before this screen shot? You know, the one that resulted in the injector pulse width doubling and flooding the engine with fuel as it failed the rationality diagnostic... I know you felt it. I can see it in the RPM and speedometer trace. Yeah, dynamic airflow is guaranteed to match the VE airflow when your experiment with the VVE and MAF cal forces it to ignore the MAF because it suspects hardware failure. I'm not surprised you never got back into the throttle after that little incident...puckered the asshole a little bit, eh?

Don't be dishonest dude. We can literally see it in the data.

What your cherry picked screenshot doesn't show is the correlation of steady state throttle where dynair is using MAF and all of a sudden...BAM dynamic air switches to VVE input and floods the motor! Tis you that is being dishonest.


Attachment 132248

[smokeshow]

Hell yeah I felt it and it scared the shit out of me for sure! But then again dude, this is a result of me intentionally jacking up the airflow model to teach the class! This does not happen when I roll around on my normal tune. That all you got?

I mean, that approach is fine. So long as you don't try to justify an incorrect conclusion with it when it does fault out. I understand getting pulled into a rabbit hole...as long as you eventually come back out better educated lol.

[smokeshow]

What your cherry picked screenshot doesn't show is the correlation of steady state throttle where dynair is using MAF and all of a sudden...BAM dynamic air switches to VVE input and floods the motor! Tis you that is being dishonest.

What? I already addressed that...

Yeah, dynamic airflow is guaranteed to match the VE airflow when your experiment with the VVE and MAF cal forces it to ignore the MAF because it suspects hardware failure.

[Cringer]

I mean, that approach is fine. So long as you don't try to justify an incorrect conclusion with it when it does fault out. I understand getting pulled into a rabbit hole...as long as you eventually come back out better educated lol.

And I have become better educated, and my quest is never done. I wish you the best on your journey, sir.

[smokeshow]

And I have become better educated, and my quest is never done. I wish you the best on your journey, sir.

You too dude. Keep up the work with the tools. I'm serious about replacing the HPT scanner lol.

[DSteck]

Y'all are nuts and doing it wrong anyway.

[TheMechanic]

You too dude. Keep up the work with the tools. I'm serious about replacing the HPT scanner lol.

Wouldn't it be nice to use an AI to self tune to certain performance levels? Like an advanced Holley system. Your only job is to make sure you don't have mechanical issues.

[Cringer]

You too dude. Keep up the work with the tools. I'm serious about replacing the HPT scanner lol.

Well I don't "count sheep in hex" so that aint hapenning! What is your main beef with the scanner? Canbus polling/lag/refresh rate too slow?

[Cringer]

y'all are nuts and doing it wrong anyway.

player three has entered the game!!!

[Cringer]

Wouldn't it be nice to use an AI to self tune to certain performance levels? Like an advanced Holley system. Your only job is to make sure you don't have mechanical issues.

So I asked ChatGPT...

How could I write an AI program that would self tune a modern EFI ECM? I would just want to plug basic mechanical specifications of the parts and allow it to monitor all sensors and self adjust as needed to maximize horsepower and torque.

Designing an AI program to self-tune a modern EFI (Electronic Fuel Injection) ECM (Engine Control Module) can be a complex task, but I can provide you with a high-level overview of the process. Keep in mind that implementing such a system would require advanced programming skills and a deep understanding of engine control systems.


1. Data Acquisition: Collect Data
To start, you'll need a system that can acquire and process data from various engine sensors. These sensors may include throttle position, intake air temperature, coolant temperature, oxygen sensors, and more. The data will serve as the input for the self-tuning algorithm.


2. Define Performance Objectives
Next, you'll need to define the performance objectives for your engine, such as maximizing horsepower or torque. These objectives will guide the self-tuning algorithm in making adjustments to achieve the desired outcomes.


3. Develop Self-Tuning Algorithm
The self-tuning algorithm will be the heart of your program. It should be designed to analyze the sensor data and make appropriate adjustments to the ECM parameters in real-time. The algorithm will likely utilize machine learning techniques to learn the optimal ECM settings based on the input data and performance objectives.


4. Training Phase
During the training phase, you'll need a set of baseline ECM parameters that are considered to be close to optimal for your specific engine and parts combination. The algorithm will analyze the sensor data collected during this phase and make adjustments to the ECM parameters to improve performance. The algorithm can employ techniques like reinforcement learning or genetic algorithms to explore the parameter space and find better configurations.


5. Feedback Loop
To continually improve the tuning, you'll need a feedback loop where the algorithm assesses the impact of the parameter adjustments it makes. This feedback can come from measuring engine performance metrics, such as horsepower, torque, air-fuel ratio, and emissions. The algorithm should continuously adjust the parameters based on this feedback to fine-tune the ECM settings.


6. Safety Measures
It's essential to implement safety measures to prevent any potential damage to the engine. These measures can include monitoring sensor data for abnormal readings, setting limits on parameter adjustments, and including fail-safe mechanisms to revert to safe settings if critical thresholds are exceeded.


7. Iterative Refinement
The self-tuning algorithm may require multiple iterations and refinements to achieve optimal performance. You might need to collect additional data and adjust the algorithm accordingly to account for different scenarios and operating conditions.


8. Implementation and Testing
Once you have developed and refined the self-tuning algorithm, you can integrate it into a software program that communicates with the ECM. Extensive testing on a dynamometer or under controlled conditions is crucial to verify the effectiveness and safety of the self-tuning system.


Remember, building an AI program to self-tune an ECM is a complex undertaking that requires expertise in engine control systems, programming, and machine learning. It's important to have a deep understanding of the mechanics involved, and to exercise caution and thorough testing to ensure the safety of the engine and other components.

[smokeshow]

Y'all are nuts and doing it wrong anyway.

Who are you again?

Well I don't "count sheep in hex" so that aint hapenning! What is your main beef with the scanner? Canbus polling/lag/refresh rate too slow?

I mean you're a good example here lol. You wouldn't be motivated to make these tools if HP Tuners already had the capability. The scanner hasn't changed much at all in the past 10+ years. Powertrain controllers are wildly more complex and generate tons more data than they used to, yet there has been next to nothing in the way of post-processing tools added to the scanner.

Lol @ the ChatGPT response. That's a pretty good answer... I've always figured that if I owned a shop and a POS Mainline like Dave's, much of the calibration would end up automated. No sense in sitting in the car to do mundane data collection when the equipment is almost fully capable of capturing speed/load data already. It was standard practice at Chrysler to set the engine dynos to begin tests near the end of the work day and let them run overnight. The mainline could do that as well.