Page 59 - Sustainability in the Process Industry Integration and Optimization

36 Cha p te r T h r ee

involve choices between structural options and involves
decisions on flow rates and capacities only, then continuous
optimizers (LPR or NLP) can be used.
2. Automated generation of the maximal superstructure, followed by
enumeration of all feasible network topologies using the P-graph
framework. The unit operations can be described in terms of
(a) the input to the automated procedure, (b) the compatible
connections between them, and (c) the corresponding process
and cost information. More details are provided in
Chapter 7.

The construction of a process network model begins with
formulation of the mass balances. The following key points should
be kept in mind:

• Mass balances can be of two types: overall and component-
wise. Overall balances are performed over the total contents
of the input and output streams of a process unit. For steady-
state models of continuous processes, this content is usually
expressed as a mass-based flow rate (e.g., in units of kg/s,
kg/h, or t/h). Componentwise mass balances reflect the mass
conservation principle between the inlet and outlet streams
for individual chemical components (or pseudocomponents).
• For a given operating unit, the set of all componentwise mass
balances is exactly sufficient for completely characterizing
the material flows into and out of the unit (adding the overall
balance could lead to an “overspecified” system of equations).
However, if using the overall material balance is critical to
the system model, then the overall balance can be used in
place of one of the componentwise balances.
• In some cases, tracking all chemical species in various process
streams is not necessary. This is true for water networks in
which mass balances are written for the water flow rates and
for the analyzed contaminants. However, an incomplete list
of the material species contained in the water streams will
naturally result in incomplete mass balances (Smith, 2005).
When complete results are necessary, rigorous simulations of
the optimized system must be performed.
• The componentwise mass balances of stream mixers and the
more complex operating units involve bilinear terms that
reflect products of the stream flow rates and the component
concentrations. In this case, the result is an NLP or MINLP
problem. If either the concentrations or the flow rates are fixed
then the model could be linear (an LPR or MILP problem),
which would make for an easier computation that might
guarantee global optimality. Although such an approach is