Академический Документы
Профессиональный Документы
Культура Документы
Abstract
In this paper, we propose a generic computer-aided methodology for synthesis of
different processing networks using superstructure optimization. The methodology can
handle different network optimization problems of various application fields. It
integrates databases with a common data architecture, a generic model to represent the
processing steps, and appropriate optimization tools. A special software interface has
been created to automate the steps in the methodology workflow, allow the transfer of
data between tools and obtain the mathematical representation of the problem as
required by the optimization tool. The methodology and its implementation have been
tested through various case studies.
Keywords: synthesis of processing networks, superstructure optimization, mixed-
integer nonlinear programming (MINLP), ontology-based data architecture.
1. Introduction
The synthesis of processing networks is an industry-relevant problem. A processing
network is combination of raw materials, a series of processing steps to convert them,
and products which they can be converted to, where each processing step has various
alternatives in terms of processing technologies, giving rise to a superstructure of
alternatives. The synthesis problem is formulated as an optimization problem, solved to
obtain the optimal network topology along with the values of continuous variables (such
as operating conditions), from a superstructure of alternatives. Due to its decision-
making nature, the synthesis problem involves a combination of continuous and discrete
variables, hence leading to its formulation as a mixed integer nonlinear optimization
programming (MINLP) problem.
The above definition of processing network applies at the same time to different cases,
such as chemical processes, wastewater networks, biorefineries and CO2 utilization
processes. Therefore, a generic approach to solving the synthesis problem in different
types of networks is proposed. The formulation and solution of this class of problems
involves large amounts of data, especially for large problems, coming from different
sources, making its systematic storage and organization essential. Moreover, these data
need to be used to generate all possible alternatives, hence calling for a superstructure
generation method. Next, alternatives are represented mathematically, through a generic
model, which needs to be able to represent all alternatives for different problems.
Finally, the optimization problem is to be solved, using an appropriate external solver.
As a consequence, the methodology needs to be systematic, generic and integrate
various methods and tools (database, superstructure generator, optimization model and
686 M.-Ona Bertran et al.
Figure 1: Problem types: (a) route selection; (b) product selection; (c) feedstock selection;
(d) simultaneous feedstock, product and route selection; and (e) feedstock and product selection
via an intermediate.
A Generic Methodology for Superstructure Optimization of Different
Processing Networks 687
The problem type depends on which materials are fixed (or specified by the problem),
and whether there are alternative routes or not. For example, problem type (a) in Figure
1 is a route selection problem, feedstock and product materials are specified, and there
is a number of alternative routes; problem type (b) in Figure 2 is a product selection
problem, feedstock is fixed, but product is not, and there is only one route per product.
2.1.2. Step 2: Superstructure generation
To generate the superstructure, one or various databases are used to retrieve all the
necessary information, that is the alternatives, connections between them, and data for
each alternative. The alternatives in the superstructure are represented as processing
intervals (Quaglia et al., 2012). Connections between processing intervals are inferred
by comparison of inlet and outlet materials of each unit in the database (see Section
2.2). If new information that is not in any of the databases is to be used, it is first input
to one of the databases, so that the data has the same structure as the rest and it can be
reused in future problems. For this work, two different databases have been created and
used: a biorefinery database (the ProBioRefine database) and a CO2 conversion
database.
2.1.3. Step 3: Solution of the optimization problem
First, the mathematical representation of the superstructure is developed. Next, the
optimization (MINLP or MILP) problem is solved. This problem includes the user-
defined objective function, subject to a series of constraints, namely mass and energy
balance constraints, flow model constraints, logic constraints, and other constraints
(Quaglia et al., 2012). The developed model is generic and data-independent. This
means that it does not have to be further edited by the user, the necessary data needs to
be provided as an input file and the problem is solved using external software. Through
the developed interface, Super-O, the input file is automatically generated by retrieving
the necessary data from the databases.
2.2. Data architecture
The relevant information for this problem includes the alternatives for each processing
step and connections between alternatives in different processing steps, as well as
detailed data for each alternative. This represents a large amount of data, which calls for
the development of a data architecture for its efficient storage, search and retrieval. The
data structure needs to be flexible enough so as to be applicable to a diverse set of
problems, yet robust enough so as to provide a strong common format for all data. This
means that data obtained from various sources and disciplines, including open literature,
academic and industrial collaborators, and reaction path synthesis, should be easily
combined, allowing the solution of assorted problems together. To this end, an
ontology-based data architecture has been developed, around which specific databases
are built. The developed ontology, shown in Figure 2, consists of two main sections,
namely a materials section and a technologies section. The materials section contains
the list of materials, and each of them is assigned an attribute (feedstock, intermediate,
or product), which creates a link between the material and one of the material sub-
sections. The material sub-sections are: a feedstocks sub-section, an intermediates sub-
section, and a products sub-section. This means that the materials with attribute
“feedstock”, appear in the feedstock list, where they are classified (into types) and given
all the necessary data. The units section classifies the units in types and sub-types, as
well as containing all information about each unit. The link between the materials and
the units section is done through assigning one material of the materials list as inlet and
688 M.-Ona Bertran et al.
another one as outlet of each unit. Therefore, connections between units are inferred by
comparison of inlet and outlet material names, hence obtaining a superstructure of
alternatives.
Material Feedstock type
M FT
Attribute
Feedstock
M1 Feedstock FT1
F1
Intermediate
Location
Product F1,1
L1,1,1
Unit Price
UST1,1
U1,1 Product type
PT
...
Unit Data
U1,1,1
UD1,1,1 Product
PT1
P1
Inlet Outlet
Mixing data
M1 M1 Characteristics
Reaction data P1,1
PC1,1,1
...
Demand
Specifications
...
Figure 2: Ontology-based data architecture (left: main sections, right: material sub-sections).
Transformations might be needed after retrieving data from the database for its use in
superstructure optimization. Data in the database is stored in terms of units, while a
superstructure is generated in terms of processing intervals. A processing interval can
include a part of a unit, a whole unit, or various units (even a whole plant). Therefore,
the definition of processing intervals and what they include depends on the problem
scale. Each processing interval is modelled using a generic model, thus the same data is
required for each of them. Simple algorithms and calculations that allow for the
conversion of units data (from the database) to the processing intervals data (for the
superstructure) have been developed.
3. Software implementation
In order to automate the process synthesis stage of the three-stage approach, the
software program Super-O has been developed in C# environment. Super-O is a user
interface for the synthesis of processing networks based on the presented methodology.
It allows for the reduction of the time needed for the formulation and solution of
network optimization problems. A schematic representation of the methodology as
implemented in Super-O including the data flow and tools integration is shown in
Figure 3.
A Generic Methodology for Superstructure Optimization of Different
Processing Networks 689
Step 1.
User specifications
Problem formulation
Problem
specifications
Step 2.
Superstructure generation
Databases
Table 1: List of case studies solved by Super-O. Abbreviations as follows: NPI is the number of
processing intervals; NF is the number of feedstocks; NP is the number of products (note that NPI
includes NF and NP); NC is the number of components; NR is the number of reactions; NEQ is
the number of equations; NV is the number of variables; NDV is the number of discrete variables.
Problem size Model size
Case (problem type*) Reference
NPI NF NP NC NR NEQ NV (NDV)
Network benchmark 12 2 4 5 2 3,476 3,235 (120) Quaglia et
problem (d) al. (2012)
Sugarcane molasses 32 1 3 12 26 76,360 73,141 (52) Bertran et al.
biorefinery (b) (2015a)
DMC from CO2 (a) 16 9 5 8 30 61,534 59,783 (5,400) Kongpanna
et al. (2014)
Wastewater network 24 2 6 15 37 112,147 108,742 (74) Handani et
(d) al. (2014)
Biodiesel biorefinery 46 3 6 27 91 1,210,227 1,193,507 (182) Bertran et al.
(d) (2015b)
*As defined in Figure 1
References
Babi, D.K., Holtbruegge, J., Lutze, P., Gorak, A., Woodley, J.M. & Gani, R., 2015, Computers
and Chemical Engineering 81, 218-244.
Bertran, M., Anaya-Reza, O., Lopez-Arenas, M.T., Woodley, J. & Gani, R., 2015a, Synthesis of
biorefinery networks using a superstructure optimization based approach, ECCE10, Nice.
Bertran, M., Orsi, A. & Gani, R., 2015b, Superstructure-based optimization of biorefinery
networks: Production of biodiesel. In M-O. Bertran, T. Bisgaard, & R. Frauzem (Eds.), Book
of abstracts for PSE-2015/ESCAPE-25.
GAMS Development Corporation, 2011. General Algebraic Modelinng System (GAMS).
Handani, Z.B., Quaglia, A. & Gani, R., 2015. Computer Aided Chemical Engineering, 37, 875–
880.
Kongpanna, P., Pavarajarn, V., Gani, R. & Assabumrungrat, S., 2014, Chemical Engineering
Research and Design 93, 496-510.
Quaglia, A., 2013a, An Integrated Business and Engineering Framework for Synthesis and
Design of Processing Networks, Technical University of Denmark, PhD Thesis., Kgs. Lyngby.
Quaglia, A., Sarup, B., Sin, G. & Gani, R., 2012, Computers and Chemical Engineering 38, 213-
223.