Air Flow Measurement

oecar1 at oec4.orbeng.com.au oecar1 at oec4.orbeng.com.au
Thu Nov 21 03:15:52 GMT 1996


>Ok, so go buy a $90 vane/rotating type meter that is 1-2% accurate 
>and get a curve for your MAF to linearize it.

I'm afraid to disappoint you, but there just 'aint such a beast.  The 
error specs for most automotive spec thermal MAFs are +/-4% of value. 
A flapper type AFM would never even come close to this.  I have used 
both a NIST traceable Subsonic Venturi and a NIST traceable Laminar 
Flow Element for my reference airflow measurement (+/-0.5%).  Each of 
these is worth upwards of $US2500 a piece.

>Heck, take the throttle body off the car (w/ the intake plumbing 
>intact), put the vane sensor on a pipe at the output (to make the 
>flow even and laminar) and stick the whole thing out the window of 
>you car and take some data points;)

I've been using an engine dynamometer.  MUCH more convenient, but the 
problem is not in collecting the data... 

>Should be really easy to find a 2nd,3rd,...nth order approx to 
>linearize the sensor (or use yet another LUT (look up table) for the 
>computationally challenged).

I have a fourth order curve which fits my data to better than 1% RMS 
error when testing at steady state flow.  The problem is that steady 
flow conditions only exist at very low throttle openings. 

As you open the throttle the steady flow starts to become pulsating 
flow and if you're unlucky it will turn into reverse flow at some point.  
To give you an idea of the magnitude of this, at WOT, 2250 rpm on the 
engine I'm working on (4-cyl) flows about 37g/s average, but between 25 
and 54 g/s peak-to-peak.  The frequency is twice engine speed for 
obvious reasons

The problems become apparent when you know that the sensor output is
non-linear, the frequency response is asymmetrical, and any reverse
flow signal will be rectified.  This means that the true average 
(voltage) signal will not represent the true average airflow.  Any 
digital approximation to this will only be worse unless you use a few
tricks.

To convert this analogue signal to a useful byte or two, you must 
perform some A-D conversion.  So you are quantising it not only in 
the range, but also in the (time) domain.  If you don't do this syn-
chronously (ie you sample in the time domain) you will wind up with 
a very noisy (and aliased) signal.  If you do it synchronously (in the
crank domain) you have to decide what angle gives you the represent-
ation of the average airflow.  Your sample angle cannot be calculated
(easily) and varies significantly with engine speed (and to a lesser
extent with throttle position)

I guess what I'm trying to suggest is that (from my experience at
least) the MAF is no slap-it-on-and-go solution.  A MAP sensor on
the other hand is linear, can be throttled to give a physical average
or can be sampled and averaged/filtered in the time domain without
difficulty.  It has it's disadvantages, but...

There's heaps more to be said about this subject but I hope it's
useful to share this with you guys.  If anyone has any other ideas or
experience on getting MAF's to work well, I'd love to hear from them.

Andrew Rabbitt
Orbital Engine Company
Perth, Western Australia




More information about the Diy_efi mailing list