Page 118 - Sustainability in the Process Industry Integration and Optimization
P. 118
P r o c e s s I n t e g r a t i o n f o r I m p r ov i n g E n e r g y E f f i c i e n c y 95
on what assumptions are adopted, it is possible to obtain both MILP
and MINLP formulations. Linear formulations are usually derived
by assuming isothermal mixing of the split branches and then using
piecewise linearization on the heat exchanger capital cost functions.
With the superstructure approach it is possible to include other
heat exchange options—for example, direct heat transfer units (i.e.,
mixing) and different heat exchanger types (e.g., double-pipe, plate-
fin, and shell-and-tube). Soršak and Kravanja (2004) presented a
method incorporating different heat exchanger types into a
superstructure block for each potential heat exchange match. Some
other interesting works in this area are by Daichendt and Grossmann
(1997), Zamora and Grossman (1998), Björk and Westerlund (2002),
and Frausto-Hernández et al. (2003).
4.5.3 A Hybrid Approach
It is clear that the superstructure methodology offers some
advantages in the synthesis of process systems and, in particular, of
HENs. Among these advantages are: (1) the capacity to evaluate a
large number of structural and operating alternatives simultaneously;
(2) the possibility of automating (to a high degree) the synthesis
procedure; and (3) the ability to deal efficiently with many additional
issues, such as different heat exchanger types and additional
constraints (e.g., forbidden matches).
However, these advantages also give rise to certain weaknesses.
First, the superstructure approaches, in general, cannot eliminate the
inherent nonlinearity of the problem. Hence they resort to
linearization and simplifying assumptions, such as allowing only
isothermal mixing of split streams. Second, the transparency and
visualization of the synthesis procedure are almost completely lost,
excluding the engineer from the process. Third, the final network is
merely given as an answer to the initial problem, and it is difficult to
assess how good a solution it represents or whether a better solution
is possible. Fourth, the difficulties of computation and interpreting
the result grow dramatically with problem size; this is a consequence
of the large number of discrete alternatives to be evaluated. Finally,
the resulting networks often contain subnetworks that exhibit a
spaghetti structure: a cluster of parallel branches on several hot and
cold streams with multiple exchangers between them. Because of
how superstructures are constructed, these subnetworks often
cannot be eliminated by the solvers.
All these considerations highlight the fundamental trade-off
between applying techniques that are based on thermodynamic
insights, such as the Pinch Design Method, and relying on a
superstructure approach. It would therefore be useful to find a
combination of both approaches and, if there is, to see whether it
offers any advantages. One such middle way is the class of hybrid
synthesis methods described next. A method of this class first