FW: DIY_EFI Digest V1 #303

[Mark Pitts]

Could you copy me on the maths side as well please.

Mark

----------
From:  William Boulton[SMTP:boultonw at OntheNet.com.au]

> >My solution is actually to use RPM(T-1) and RPM(T) to calculate 
> >RPM(T+1) :
> >  RPM(T+1)=RPM(T)-RPM(T-1)+RMP(T)=2*RMP(T)-RPM(T-1)
> >

> >This give good results (much more better than using RPM(T+1)=
> >RPM(T))
> >
> >Maybe somebody know a better solution (Kalman Filter
> >for instance, or ...) ?
> 
> You could try RPM(t+1) = 3 x RPM(t) - 3 x RPM(t-1) + RPM(t-2)
Like to add 2cents to the debate.  The best way (IMHO) to achieve any
degree of accuracy with timing prediction is to run as many reference
points per revolution as possible. I've see the result of 1/rev on a twin
and that was a disaster. Engines just do not maintain consistent angular
velocity during a revolution. I developed a simulator some time ago to
test theory on this and found that good results are obtained with at least
6 reference points per rev. I tried a number of basic equations and
settled on a variation of the one Chaxel used. Just multiplied the result
by some large fraction just less than 1. Still easy in integer assembler.
The problem in predicting timing stems mostly from dA and not dV.

Paul, I'd be interested in how you arrived at your equation. RSVP

Bill Boulton.