Page 349 - Design of Simple and Robust Process Plants

8.4 Control Design 335
The search for such a structure is of value for unit control, and it is often difficult to
close two quality loops (such as for a distillation column) without introducing severe
interaction that would require a model-based controller. On the other hand, it is
quite common that only one of the objected control values is a hard constraint like a
product specification, while the other is economically determined ± a so-called soft
specification. A third point is that steady-state operation optimizations set points are
implemented over a time period of hours or even days for off-line optimizations. In
the meantime, disturbances might be introduced which must be absorbed. In these
intermediate times the units still need to be operated at close to optimum condi-
tions. The above points were a reason to search for self-optimizing control structures
of individual units, reactions and separations, with minimal losses as a consequence
of disturbances while limiting the amount of interaction. A typical example of a self-
optimizing control structure for a distillation column is a top and bottom quality
control. Such a structure however does not comply with the requirement of limited
interaction, which is basic for a simple and robust process plant.
Skogestad (1999) presented a step-wise methodology for development of self-opti-
mizing control structures by comparing different options, and illustrated this with
examples for reactor and distillation control.
The steps for the distillation example Figure 8.17 (a propylene/propane splitter
with 110 theoretical stages) were:

1. The selection of the basic design conditions for reference; the distillate spec
x D is considered as a hard constraint 0.995.
2. Selection of disturbances in the example were feed flow increase of 30%, feed
composition z F from 0.65 up to 0.75 and down to 0.5 parts of lights, decrease
in liquid fraction q F from 1.0 to 0.5 (50% vapor) and a impurity increase of
distillate from 0.995 to 0.996. Economic disturbance was reflected in the
price of energy, which was increased a factor five from 0.1 to 0.5 $/kmol boil-
up V, the price p for distillate 20 $/kmol, which was increased to 30 $/kmol
in reference to a feed cost of 10 $/kmol.
3. Selection of the potential controlled variables.
4. Calculation of the steady-state economic optimal conditions of the distillation
for the selected variables and the different disturbances.
The results for the optimized distillation conditions in respect to the selected distur-
bances are shown in Table 8.4 including the nominal conditions. Some of the con-
clusions were:

± The values are insensitive to the feed rate (as was to be expected), as the effi-
ciencies are not capacity-dependent, and the feed-forward action implemen-
ted for R, D, and V.
± Optimal bottom composition is rather constant, with the exception for the
high energy price.
± Optimal values of D/F varied significantly, which seemed to be a bad choice
for a CV.

Based on the results, candidate CVs were selected.