[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: '747 and Proportional Stuff



On Tue, 11 Jul 2000 16:19:10 -0400, rr <RRauscher@nni.com> wrote:

>Just try to sneak something in, and I've got questions on it <g>. That'll
>teach me. . . lol

Hey Roberto, hey doods.

I've been too busy to insinuate meself into the thread till now, but I
have to say I think this is very HOT stuff. I wonder if there'd be any
interest in trying to concoct a 'keeper thread' on the actual PID
algorithm.

Well, lemme say I'm sure that's what's up here already pretty much with
that excellent lil 747Prop.txt file you uploaded, but I'm just thinking
if we could get you to post some of the code loops we might be able to
flesh out the makings of a WOT PID setup. You know my undying interest
in THAT subject, o'course. :)  Think of it, if that could be made to
work, the price barrier to WOT C/L controllers would crumble just like
Jericho. Not to mention what it would do for the EGORmodule market. ;)

Also, I've never heard of anyone publishing anything on the actual
discrete-time PID algorithm used in these bang-bang controllers, has
anyone? I don't mean a college text on control theory, but a tutorial on
the actual inner workings of the ECM's C/L code-loop from a basic
PID-technical standpoint.

But first things first, I'd just love to understand the current "limit
cycle" PID algorithm that you're dissecting at present. My EE training
idn't in that end of things sadly, but I've been doing some reading on
PID, and found your "Derivative effect = resetting of PropGain & Lag"
MOST illuminating. Was at a loss as to where the "D" in PID could
possibly be gotten. I'm gonna have to think about it some more, but it
sounds right.

So here's me question/request, which unfortunately as an ecm-challenged
dweeb, I can only tug at, but can't do much meself. Instead of the
actual code and all the difficulties/details of getting into that, could
we start putting together (there's that theatrical "we" shit again :) a
"psuedo-code" of the PID algorithm (you know, basic equations & steps in
the loop), that you're studying now in that 747. To the point where the
timing of the loop could be estimated, especially.

I *could* offer THIS much contrib, if we could figure the duration of
the PID control loop (ya know, how fast each computation cycle/loop is
in uSecs), I might be able to whip together an Excel model/simulator, so
we could experiment with the PID tuning that way. Not only might this
help in everyone's understanding of the dynamics of C/L, but we could
tweak things like the deadtime from exhaust to O2 sensor, the response
time of the O2 sensor, etc. etc. and see what effects this is likely to
have on the reaction time and tightness of the PID loop in the 747.

That might just be all the critical mass we'd need to get some ideas on
how to put WOT C/L into orbit (I have a sneeking suspicion you might
already HAVE some ideas on that, heh :). At LEAST, it would allow us to
evaluate the possibility of using the conventional limit-cycle/bang-bang
control algorithm with this idea that's been buzzing around ever since
talk of WB O2s got started; that is the idea of faking the
"stoich-crossing" point to the OEM controller (whilst holding it in
C/L). This at least might allow us a piecewise WOT controller, that say
gives us stoich if being smogged, nice smooth slightly rich idle
otherwise, plus an adjustable lean cruise, and a nice fat 12.5 at HO. As
has been already discussed, we might be able to detect these "regimes"
pretty easily outside the ECM, and switch the "stoich-crossing" around
to get those "piecewise" control points. But the real question is, what
kind of PID control performance could we get from just using the
conventional limit cycle (and thereby avoiding major code changes);
that's where a decent simulator would be good, because even with on-car
or even bench test setups, it's hard to "schmoo" variables like
deadtime/transport delay and O2 sensor response time to see where the
"sensitive spots" are.

I dunno, given the way the conventional INT & BL cells are used, the
above might simply be a pipe dream, and need some significant code
changes anyway, but starting with the algorithm in some form that could
be tested/simulated in EITHER case I think would be a good thang.

Just some itinerant synapses...

Gar


----------------------------------------------------------------------------
To unsubscribe from gmecm, send "unsubscribe gmecm" (without the quotes)
in the body of a message (not the subject) to majordomo@lists.diy-efi.org