Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!uunet!rosevax!bert.Rosemount.COM!bill
From: bill@bert.Rosemount.COM (William M. Hawkins)
Newsgroups: comp.realtime
Subject: Re: Data acquisition/control on i386
Message-ID: <8461@rosevax.Rosemount.COM>
Date: 5 Jan 90 04:48:16 GMT
References: <1989Dec10.094755.21999@me.toronto.edu> <1989Dec28.104004.5697@me.toronto.edu> <1990Jan4.074652.1026@stag.UUCP>
Sender: news@rosevax.Rosemount.COM
Reply-To: bill@bert.Rosemount.COM (William M. Hawkins)
Organization: Rosemount Inc., Burnsville, MN
Lines: 33

In article <1990Jan4.074652.1026@stag.UUCP> trb@stag.UUCP ( Todd Burkey ) writes:
>In article <1989Dec28.104004.5697@me.toronto.edu> kokody2@me.utoronto.ca writes:
>>Also of interest is
>>how robust the data acquisition/control is under varying machine loads.
>
>Very robust, since our controllers use an extremely tunable analog PID
>configuration.

Todd, what do you mean by "extremely tunable"?  It has been my
experience that PID's are not tuned to constants (P, I, and D)
with more than a few significant digits, because the real world
conspires to change the gains and time constants of electro-
mechanical devices.  The Nyquist stability criterion is usually
met with some safety factor, called phase margin.

The original question dealt with stability under varying machine
loads.  This becomes a problem when the dead time between taking
a measurement and manipulating the output is more than 10 percent
of the dominant lag time.  If the external process has transport
lag, add that to the processing time.

The sampling and controlling interval is not dead time.  It can
be a problem when it is more than 25% of the dominant lag.  That
is, it is not dead time unless you wait until the start of the
next interval to send the outputs from the last interval.

bill@bert.rosemount.com

P.S.  We make process control systems that usually operate
pneumatic valves from 1/2 to 48 inches in diameter.  We use
Motorola 68xxx processors, with a proprietary operating
system written in C under UNIX.  It is optimized to minimize
delays in handling clock tick interrupts.