Posted by Dan O'Connor on November 09, 19101 at 17:52:53:
In Reply to: Re: Mixed process dynamics in DMC posted by Adrian Meaburn on August 30, 19101 at 12:29:22:
Adrian,
You are correct, mixed processes do exist, but in my view, I probably
would still call the mixed process variable a ramp variable, because
it is likely that some of the MV's cause it to integrate. An example
of such a process is a C3 splitter (Propane/Propylene), where the feed
rate is on FIC control, the bottoms flow is on FIC control, the bottoms
level is cascaded to the reboiler, the top reflux is on flow control, and
the overhead level is open loop, with the top product draw on flow control.
In this system, if you move the top product draw, the overhead
drum level behaves as a ramp. If you move the reflux, the overhead
drum level behaves as a steady state variable, or if you move the
pressure setpoint, it behaves as a steady state variable. To me, the
level is still a ramp, and should be identified as such, but the
rate of change at steady state should be zero for the associated ss variables. I have found that
the new subspace identification in DMCPlus MODEL works nicely with
ramp variables. However, there seems to be a bug if you try to
run the identification on a mix of ramp and ss dependent variables.
If you have MODEL 3.1, try out an identification with the variable
as a ramp, but set the detrending filter time to 9999, and the
oversampling ratio to 1. Then try the same analysis with it as a
steady state variable, with the detrending time to 9999, oversampling
to 1 - you'll find it gives the same answer, and that the identification
is less susceptible to noise that the DMI, because I'm pretty
certain it's not differentiating the data to identify the ramp.
Hope this helps.
Sincerely,
Dan
I inherited some RMPCT models for a reactive distillation column which
: appeared to exhibit mixed dynamics. I am now retesting the plant as I am not
: sure about them. The root of the problem is that the process is very slow
: and is very difficult to test effectively using standard steps (i.e. 24hr
: testing). Preliminary data from the prbs suggests that the original models
: were flawed and the process does not contain these mixed dynamics.
: The impression I get is that mixed dynamics are not a common occurrence on
: process equipment. However, I have seen this mixed behaviour once before on
: another of our process process but have not had to deal with it directly as
: the DMC controls work well when separated (a legacy of RMPCT). This is a
: process consisting of a stripping column attached to a reactive system with
: large recycles. I am confident of the models as the controls work very well
: but unfortunately cannot go into too much detail as this process is
: proprietary.
:
: Adrian