Experimental Physics and
Industrial Control System

Brian Bevins <[email protected]> · Thu, 15 Jun 2000 10:33:13 -0400

Leo Dalesio wrote:
> 
> It seems that we all agree that the one in base needs to go. Perhaps we
> could include both EPID and CPID. Rename EPID to be PID. It appears to be a
> correct version of what PID attempted to do. CPID is special in that it
> includes some model features. If CPID has some of the same problems that
> were found in PID, it would be worthwhile to fix them. Someone at JLab may
> want to look at this.
> Is anyone using the original PID? SUccessfully? That would mind if we
> changed this?
> 
>     Bob

At JLab we have never run into the problems Mark describes. The problem
of output saturation only arises if you use the PID as a
proportional-only controller (KI=0,KD=0) and we never do that. I'm not
sure I understand why you would ever want to do that, since the PID will
never reach its setpoint that way. If you use an appropriate KI term,
the behavior of both loops becomes identical and saturation is not an
issue. (I have developed a spreadsheet similar to Mark's that
demonstrates this which I will put up on the web if anyone is
interested.)

The second problem of integrator windup is not an issue for either cpid
or pid since they only calculate their output based on the first
difference of the PID expression. Once the output becomes saturated, no
further integrator windup is possible. This results in the same behavior
as the epid record limiting the I term to stay between DRVH and DRVL.

If I am missing something and there are good reasons to want to use a
PID for proportional-only control, perhaps Mark and I could combine the
best features of cpid and epid into a single new pid record.

Cpid is fully described in the R3.13 Record Reference Manual at:

http://www.aps.anl.gov/asd/controls/epics/EpicsDocumentation/AppDevManuals/RecordRef/Recordref-16.html

There are features in the cpid record that I think would be generally
useful and that I would definitely like to see in any new pid record
included in base. 

1) Built-in support for a direct readback from the controlled device
(e.g. motor speed, valve position) in addition to the input signal used
for control (e.g., flow rate, liquid level).

2) Can function in integral-only mode (pid and epid can't). Cpid
actually uses the PID expression shown in the Record Reference Manual,
pid does not. We don't use them this way, but some people do.

3) The MAX and MIN fields limit the range of the record output (like
DRVH and DRVL in epid) and the DMAX and DMIN fields limit the rate of
change of the output. Calculated changes smaller then DMIN are not
written out.

4) Has separate modes for normal operation, manual override, sequencer
override, and local device control and makes smooth transitions between
modes. These are essential to the way we use the cpid records. Among
other things they allow for "bumpless" ioc reboots without shutting down
the equipment controlled by the pid's.

5) Has two output modes: "Change" and "Position". Change mode works like
the old pid record and writes out only the changes. Position mode works
like epid and writes out the absolute output value. We use position mode
for devices that position absolutely (like pneumatic actuators) and
change mode for devices that position relatively (like servo motors).

--Brian Bevins
  Accelerator Controls group
  Thomas Jefferson National Accelerator Facility

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System