Re: [CAD_CAM_EDM_DRO] Re: Lost Steps => time for microprocessors!

ballendo@... wrote:

>
> Ron,
>
> Using an outboard processor is the time-proven way to do cnc. But
> even some of the big boys are using the PC itself for motion tasks.
> In effect, the PC IS THE OUTBOARD PROCESSOR. Look inside modern
> commercial CNC machines, you may find 2 or 3 PC motherbds! One for
> motion, one for GUI/ network, maybe one for IO.
>
> >Why do we insist on having the PC do the pulses? The PC ought to do
> >GUI, Math,Lookahead, etc.<snip>
>
> Several reasons come to mind:
>
> Cost = Lower (obvious)
>
> Maintennance = Easy (tens of THOUSANDS of programmers familiar with PC
> / no extra hardware to fail/ troubleshoot/ service.
>
> Manufacturing considerations = Less potential for problems/ rework/
> third-party vendor delays/ warranties on addt'l hardware/ Parts
> availability or change of spec /
>
> Natural Progression of technology = PC's faster all the time, so its
> getting easier!
>
> Versatility and Improvements = Change the code / change the product
> capability
>
> Open architecture = Exactly the problem you mentioned w/Fashcut is
> avoided
>
> There are other reasons, but these show some "whys"
>
> >Yes, Linux has a 'real time'capability <snip>
>
> Linux's real-time capability rests on the standard PC hardware!
>
> The only real limitation at this point is the ingenuity to use the
> parts of the PC system WELL. As you know, there is a tendency towards
> "this is how its done" or "It's always been done that way". But, our
> platform (the PC) is changing. The pentium RSTCD instruction (not
> sure i've got that code exactly right) gives us timing resolution AT
> THE SPEED OF THE PROCESSOR! Just that not everyone has pentiums, yet.
> The 8254 timer can be (currently, this is how RTLinux is done)
> programmed to 838ns. Interrupt latency adds to this, but EMC shows it
> CAN BE DONE.
>
> There ARE VXD's (virtual device drivers,used by win9X,NT,and WinME)
> available which support "real-time" windows; They are just VERY
> expensive.
>
> I think Art Fenerty's "Master" program is on the right track. Maybe
> "Kcam" also (kellyware.com) but I don't know kcams' internal
> architecture. They both need some polish for sure, but they are
> taking advantage of the points listed above.
>
> Also, JonE's msg is right on the mark, I think. If you're going to
> leave the PC to the tasks you mentioned, then let's use a low-
> cost,simple approach capable of hi-speed. Step-rates are increasing
> all the time (microsteppers, and servos are driving this)
>
> And,the cybernetics Cy545 chips someone mentioned were $75/axis!!! in
> 1994! Other motion chips and boards work extremely well , but again,
> they cost arms and legs.
>
> I would love to see you put your programming time into using (well)
> the PC parts we have. With your UNIX background (from earlier posts)
> I am sure you could advance the state of the art!
>
> Ballendo
>
> P.S. One area that seems to be left behind in modern programming is
> the use of some of the "old tricks" used when PC's were slow, narrow
> bitted, memory poor creatures. We just code up the high math and let
> the incredible speed of the processor "make up for it". I believe
> that the use of some of these "old tricks" will free up enough
> processor time to EASILY accomplish it all on one PC!

When speed is important; find/code a better algorithm.
Multiple devices especialy Jon's fpld sound like the right approach.

Dave