Page 381 - Design of Simple and Robust Process Plants

368 Chapter 9 Operation Optimization
mized transient. In the case of rapid dynamic effects, steady-state detection is not
applicable.
Several methods are applied to steady-state detection, including:
± mean value of a given set of measurements over a defined time period are
within a certain tolerance.
± rate of change of a variable over a time period.
± constancy of time series coefficients.
An example of a steady-state detector is the application of exponential smoothing
filters which in a digitized form is represented as

X fi =F l .X i + (1±F l )X fi±1 (4)
X fi is calculated filtered value for history i
X fi±1 is filtered value calculated for the previous history (i±1)
X i is raw unfiltered history value i
F l is filter factor
Measurements signals have high-frequency and low-frequency noise. The high-
frequency noise (ªwhite noiseº)is caused by the instrument, while the low-fre-
quency noise is caused by process noise. It is the process noise in which we have an
interest for steady-state detection.
The process noise is calculated by subtracting a heavy filtered signal (absorbs
high-frequency and low-frequency noise)from a low filtered signal (high-frequency
noise):
Process noise = heavy filtered noise ± low filtered noise.
A tolerance will be set at the process noise over a number of values in history. If
the tolerance is superseded, the process is unsteady.
The criteria for steady state should not be too strict so as to avoid there being only
a few opportunities left to perform a CLO.
The final task is to define the criteria for decider 1:
. Is the process still in its operational optimization step?
. Is the process meeting the steady state criteria?
. Are the corrupt measurements excluded?
The overall decision is to proceed to the next sequential step ªyesº or ªnoº, based on
defined criteria. Another decision is to exclude incorrect measurement data.

9.5.2.2 Data reconciliation (DR) (Jordache and Narasimham, 1999; Kane, 1999)
CLO projects apply different methods to achieve a close agreement between simu-
lated and actual operation. The objective of this module in its reduced approach is
to:

. estimate the feed and its composition by the process system;
. detect any gross errors in measurements involved the reduced model; and
. measure the actual plant performance in terms of money.