john banks

At 6000 RPM and 1 bar boost (2 bar absolute) the airflow from the airflow meter readings and calibration tables is only about 10 times the airflow at 800 RPM and idle vacuum (0.35 bar absolute).

This is surprising since you would expect 6000/800 * 2.0/0.35 = 43 times the airflow before accounting for changes in VE. Indeed the injectors are working this sort of factor of flow harder.

I find this puzzling and it makes a MAP*RPM to MAF transfer function difficult to calculate!

The same applies in the midrange - where you get most torque and therefore you would expect highest VE, the difference in fuelling is not as much as you would expect as VE at idle should be lower.

What is going on that the airflow meter is picking up so much flow at idle but the speed density method doesn't see? It can't be temperature for this order of magnitude of error.

Comments please.

pantelis

Whooooooooooooooooooooossshh - straight over my shiny head!

beryllium

MAF signal is not linear,i believe (on MY 99 i am sure)

Bob Rawle

Hi John, do you mean 200 kpa absolute, I think you will find that the air flow itself is non linear as losses vary as velocity increases ... speed-density rules ... lol, not really but you can see why a maf has to be callibrated to the "system" you are going to use.

As a thought, stop trying to get it close to what existed b4, start mapping to what you have now, correlate it to boost (absolute), thats the "true" factor that influences det activity the most. After all its just another way of picking a map point.

cheers

bob

john banks

Thanks Bob, think I need to start looking at some Link fuel maps and see what tips I can pick up. I think your suggestion is a good one. For timing it will be easy as I just need to view map access and see which bits it is hitting. For fuelling it will be a bit more tricky as the value required in the map also depends on which load column it is hitting, however, lots of interpolation and logging along with careful attention to closed loop air fuel correction should help a lot.

Beryllium, the MAF calibration is non-linear (is almost logarithmic in the mainly used areas), what I am saying is that despite that the results still do not follow the pattern I expected. Unfortunately we do not have the details of the calculation the 99/00 ECU uses to make injector pulse width, just a few hints and lots of data.

Bob Rawle

Now that you have a system that works it should be possible to characterise it, bit like maf callibration but if you get your system doing something sensible then maybe adjust the ecu callibrations to suit it, might be a better option in the long term, the maf calibration table plus the load axis scaling might give enough scope ?

Sounds as though its nearly there.

cheers

bob

john banks

Thanks.



Interestingly the new age MAF calibration does behave more as I expected with very low load values at idle. I think there is something going on that is different in load calculation between the old and new ECUs here.

Tone Loc

iTrader: (1)

John,

Before switching to the Link i spent a bit of time trying to calibrate the MAF scaling to my induction system. Below is the data from my pre97 ecu (with my adjusted valves shown to).... looks very similar to what you posted above?



Not sure if that helps at all.

Tony.

john banks

Thanks Tony, needs a login to see. Can you email it to me johnbanks@dsl.pipex.com

I think the MAF calibration in the 99/00 ECU does not give an output which is a linear relationship to load (as in air charge per cylinder) like some other ECUs. I am plotting it out and will do a bit more later to work it out.

jim litten

I know I'm hijacking the thread, but on the MY99 (even earlier I don't know) what is the brown solenoid looking thing in series with the MAP sensor?


Jim

Pavlo

presure exchange solenoid, it allows the ECU to measure atmospheric pressure by isolating the map sensor from the manifold pressure.

paul

jim litten

Ahh! well won't go messing with that then!


ta


Jim

XNWRX

Just a suggestion John, from what I understand, but I might be wrong :
The MAF signal is supposed to be used by the ECU when it operates in open-loop. In closed-loop, fuel mixture is performed from the O2 signal rather than from the MAF signal.
The MAF is accurate only from a certain air flow which corresponds to what the engine produces at about 2000/2500RPM (starting point of the the linear part of the MAF output voltage vs air flow). Below this air flow, the MAF is innacurate, but its not a p^roblem as it is not used for fine tune by the ECU in this case.
You should perform your test again, comparing what happen between 2500RPM and 6000 RPM. That time it should be coherent.
cheers

john banks

Thanks for the input, trouble is the MAF curve is never linear, and is actually very accurate below 2500 RPM. If you run it open loop below there with the MAF sensor in you can get just as accurate results as you do over 2500 RPM. Typically lambda corrections are only a few % either way and not enough to explain this discrepancy in calibration.

I am convinced that the output of the 99/00 MAF calibration table is not actually proportional to airflow but has another function on it which the ECU takes into account in its load calculation. I just need to crunch some more numbers to work it out. The alternative is a 16*16 map to look up RPM and MAP and spit out MAF. I already have a Unichip map from EMS which I have converted and would work fine, but I want to try to reduce it to a few equations to model it as these can more easily be adjusted en-bloc with a potentiometer or similar.

Bob Rawle

Well I've always advocated that, recently its become even more clear that the MY99-00 is just an early iteration of the later MY01 on ecu structure so start thinking about it like that. For example it has ign adv multiplier and ign learning which is v similar, very much faster to react though. So the maf output will_be_modified to achieve load factor. But does it matter, as long as you characterise the output that is translated into load and scale accordingly then its still just a way of picking a point on a map.

cheers

bob

Tone Loc

iTrader: (1)

John YHM.

Tony.

Pavlo

the MAF does more than just pick a point on the maps load axis. It is still used to calculate the injector duration based on the target AFR in the map. This is why any composite MAF signal needs to encompas a certain range, and give a certain resolution.

Paul

john banks

Thanks guys.

I know what you mean Bob, the fuelling is the PITA though keeping a nice shaped fuel map, and I want to end up with something that still is based on 14.7:1 as the base value, it also keeps the familiar shape of ignition maps I know which work so well, as well as all the compensations in the right places as Subaru intended. I am pretty nearly there. There appears to be an odd transformation going on for some reason.

If I prove the transfer function it is nice not only because of the shapes and contents of the maps being preserved, but also because I can then use global adjustments with temperature and an overall trimmer potentiometer in the cabin.

Also thinking laterally, a simple to calibrate DIY device that costs about £10 to replace your AFM without having to remap your ECU would be interesting to some.

[Edited by john banks - 1/4/2004 8:19:44 PM]

Pavlo

John,

I have sent you £50 by first class, can you please send 5 of said units to my home address please.

Paul

pat

A few points regarding the MAF system on the early ECUs (which may or may not correspond to the MY99 ECUs)... this is direct from a disassembly of the control code, and from my dodgy memory... caveat lector

The ECU reads the current Airflow voltage (available as QA on the SelectMonitor), it then uses this as an index to a 64 entry MAF Voltage > Actual Airflow table. This value is not available via the SelectMonitor. The actual airflow is then multiplied by the engine period (same as dividing by the RPM) and the flow scaling constant to get a per cylinder engine load (available as LOAD on a SelectMonitor). This is a very important parameter as it is used as a load index into the fuel correction and ignition timing tables. It is also inextricably linked to the fuel injector size; if you use bigger injectors then the load for a given RPM and airflow will change, necessitating either a rescale of the load axes to maintain the same timing, or for the entries in the table to be altered accordingly. Basically the LOAD parameter *IS* the base fuel injection period by another name. Of course the actual period is modified by such things as the fuel table, the warmup and acceleration enrichment, the closed loop lambda etc.

Perhaps the reason that John only saw an order of magnitude difference between idle and at full boost is that it was the computed LOAD, not the actual airflow? You'de expect to see about 10 at idle and 110-ish on boost on an early car, and about half that on the MY99/00 which is an order of magnitude. If you now multiply by the RPM again to get actual airflow then you'll see another order of magnitude.....

However this is where it gets interesting... the LOAD is an indication of fuel injection pulse width, and it is also true that you only usually see an order of magnitude difference; typically idle pulse width will be around 1.6 to 2ms and full load pulse width will be around 16ms; it is possible to run longer pulse widths but it doesn't do the injectors any favours. The actual two orders of magnitude difference in fuel delivery is due to the injectors waiting an eternity for the crank to turn round at low RPM... they squirt for the same amount of time, they just do it less frequently.

I would suggest that if trying to do a MAF to speed density conversion outwith the ECU it would be sensible to copy the original MAF Voltage -> Airflow table (from Flash99) and use this to compute the correct simulated airflow signal for any given operating point; you can include a VE correction if you want, or just do that in the actual fuel table.

Simply multiply engine RPM by the MAP and by the displacement and hey presto you have the theoretical airflow (though the units may need some frobbing). Now use this to lookup what the MAF would output for this airflow and output that from the converter.

For my car I just frobbed the ECU control code to discard the "multiply by engine period", fed a MAP signal into the MAF, made a MAP -> LOAD table to replace the MAF -> Airflow table and hey presto, speed density... now I just gotta massage the fuel table to take into account the VE, but that will be done automagically using data captured by the iPaq (when I get round to it).... eventually the iPaq will be able to update the tables live on the ECU when I get the MRAM board working

Cheers,

Pat.

john banks

Thanks Pat, your second paragraph is also my understanding. My surpise came to see that the mass air flow sensor calibration table in the 99/00 ECU does not seem to be calibrated for anything directly proportional to airflow since there is only one rather than two orders of magnitude between idle and full power, unlike it seems the earlier and later ECUs. The MAF voltage hardly ever goes significantly below 1V on the 99/00 ECU (unless you have ugly MAF flow reversions through a stalling compressor wheel or a recirculating dump valve because of air mass in the FMIC on lift off). However, at 1V there is still a significant "load" output from the MAF calibration table - even when the engine is off it is still at about 1.00 to 1.04 V, so the load should be zero if it works as expected so there is something else going on.

I have my simulated airflow ready to put into the ECU, I need a scaling for it though - not as simple as just reversing the MAF calibration table on the 99/00 ECU though, although I think it would be on the earlier and later ECUs as their mass airflow calibration tables seem to spit out airflow proper not this weird thing I have.

At a typical idle MAF voltage of about 1.3V depending on year, the early cars and the late ones both seem to read quite a low airflow - (say) 500-1000 out of a possible 65535, whereas the 99/00 is on about 5000!

I'll work a bit more on the transformation from calculated airflow to 16 bit lookup to voltage and post further findings - I have a mountain of data to look through, including Mark's working Unichip map for this function.

john banks

Looks like it will give a fairly close fit with a worst case of 10% error at medium airflows from the MY01 calibration by just shifting the MY00 calibration table along. I wonder if the throttle position sensor is having a role here.... the MY00 "airflow" calibration actually being most divergent from the MY01 at low flows when the throttle is closed and approaching it very closely at high flows when the throttle is open.

If the output of the calibration table in the MY00 is divided by RPM and then added to TPS (scaled of course) to give load it would work out pretty close I reckon. We already know that TPS has an input into load on the 99/00 ECU.

john banks

Stephen Done to the rescue

After I told him of the situation he searched the code and found that there is a &H124E offset on the output values of the MAF calibration table.

john banks

This is quite a close fit to the MY00 UK turbo MAF calibration.

I had to subtract the &H124E offset from the MAF calibration table in this model.

I will be using this to convert my calculated airflow from MAF and charge temperature into a simulated MAF voltage.

Unfortunately to convert airflow to voltage it needs to be 5th order polynomial to look any good.

I think I will recalibrate the MAF table in the ECU to a simpler function - like just taking the square root of airflow to get voltage to give more resolution at low loads. Now I know the offset and how it calculates injector times I think this will be OK.



[Edited by john banks - 1/6/2004 11:02:09 PM]