Page 47 - Sustainability in the Process Industry Integration and Optimization

24 Cha p te r T h r ee

box is a model that incorporates a physical representation, although
some of the physics is approximated—see, e.g., Hangos and Cameron,
2001). This approach is appropriate for optimizing complex systems.
If time variability has to be accounted for then steady-state modeling
can be applied to a set of operating periods, each of which is charac-
terized by its own fixed parameters.

3.2 Model Building and Optimization: General
Framework and Workflow
A good process model should contain a thorough conceptual
description of the involved phenomena, unit operations, actions,
events, and so forth. Usually this description involves text, flowsheets,
and structural diagrams. Additionally, IT-domain diagrams—for
example, UML diagrams—can be used (“UML” is a specification of
the Object Management Group; UML, 2010). The UML diagrams
include class, object, package, use case, sequence, collaboration,
statechart, component, and deployment diagrams.
A good process model should also contain a sufficiently precise
mathematical description. The mathematical relationships are used to
reflect not only physical laws but also technological constraints and
company rules. Mathematical models include algebraic equations of
some form (i.e., equalities or inequalities) and may be supplemented
with dynamic modeling, which uses differential equations to capture
variations in time, as well as states and actions to express operational
procedures and other dynamic relationships of algorithmic nature.
Structural information is also an essential feature of process network
models. When translated to Mathematical Programming (MPR)
models, such information is expressed by integer (mostly binary)
variables. An efficient alternative to representing superstructures
with binary variables is the P-graph and its related framework
(Friedler et al., 1992b), discussed in Chapter 7.
An efficient computational implementation of the mathematical
description may take the form of a stand-alone compiled application
(e.g., PNS Editor, 2010) or may be modeled within a popular
environment for process and mathematical calculations. Examples
include MATLAB (MathWorks, 2009), Scilab (2009), simulation and
optimization tools tailored for the process industry (AspenTech,
2009c), Modelica (2009a; OpenModelica, 2010), Honeywell UniSim
(Honeywell, 2010), and the open-source DWSIM (2010). All model
components have to be well synchronized to provide appropriate
user interfaces and sufficient visual aids to help understand the
process and the optimization results.
Models often include only the computational implementation
with some mathematical descriptions, but a much better practice is to
start with the concepts before deriving the mathematical relationships