Page 409 - Design of Simple and Robust Process Plants

396 Chapter 9 Operation Optimization
Some (but not all)of the modeling errors include estimation of physical proper-
ties, clustering of components, parameters depending on load (e.g., heat/mass
transfer coefficients), fixed parameters need to be determined over time, constraints
ill-defined, inertia not foreseen, nonideal flow conditions, nonideal separations, etc.
After revalidation of the unit models the overall process model is to be validated.
One of the major concerns at that point is that of the main process streams. Errors
in overall streams often do not show up to that extent in the unit models. An exam-
ple is a partial condenser splitting a major mixed hydrocarbon stream, which
depends heavily on composition, temperature, physical properties and the efficiency
of phase separation. A small error may send a different flow and different composi-
tion to another section.

9.6.12.2 Smoothing ofthe model
This step will be executed partly in parallel to the gross modeling error detection. At
this stage, we will perform off-line simultaneous DR&PEs with an extended set of
measurements and parameters, including those which were planned to be fixed in
the validated model. Kinetic constants for reactor models might also be freed up for
estimation, to obtain improved mapping between the simulation and the actual
plant operation. With the collected data sets, iterations might be executed between
model corrections (updates)and updated parameters to obtain the best fit.

9.6.12.3 Overall model validation
The overall validation is to be concluded with the performance meter (profit
meter). The performance meter is based on mass balance reconciliation and the
values of the different input and output streams. The results are expressed as a
process performance in money per unit of time based on raw material, product
and energy prices. It was concluded at an earlier stage that the profit meter needs
no validation of the underlying model as it is based on the conservation laws. The
noise of the profit meter is an important piece of information as it gives the devia-
tion in the measurement due to measurement errors and process (un)stability
(process noise). In case the spread is too large, this most likely will be caused by
process stability, with an exception for profit meter design errors, (see
Section 9.4).
There are several options to improve the spread of the performance meter due to
process instability:

. Apply a larger integration time.
. Sharpen the stability criteria for decider-1.
. Improve the control.
. Include hold-up variations in the design.

The results of the mass balance reconciliation are also used to determine the flows
and composition of the feed streams for the simulation. The overall simulated pro-
cess performance can be calculated from the process streams at the same stream
values (prices); this calculation should be part of the output of the optimization. The
spread of the simulated performance is measured for the same price set as for the