See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/262602086

Chemical Process Modeling in Modelica

Conference Paper · September 2012

DOI: 10.3384/ecp12076955

CITATIONS READS

5 414

2 authors, including:

Ali Baharev
University of Vienna
20 PUBLICATIONS   89 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Decomposition methods for global optimization View project

All content following this page was uploaded by Ali Baharev on 26 May 2014.

The user has requested enhancement of the downloaded file.

 Chemical Process Modeling in Modelica
Ali Baharev Arnold Neumaier
Fakultät für Mathematik, Universität Wien
Nordbergstraße 15, A-1090 Wien, Austria

Abstract model creation involves only high-level operations on

a GUI; low-level coding is not required. This is the
Chemical process models are highly structured. Infor- desired way of input. Not surprisingly, this is also
mation on how the hierarchical components are con- how it is implemented in commercial chemical process
nected helps to solve the model efficiently. Our ulti- simulators such as Aspen Plus R
, Aspen HYSYS R
or
mate goal is to develop structure-driven optimization CHEMCAD .
R

methods for solving nonlinear programming problems Nonlinear system of equations are generally solved
(NLP). The structural information retrieved from the using optimization techniques. AMPL (F OURER et al.
JModelica environment will play an important role in [12]) is the de facto standard for model representation
the development of our novel optimization methods. and exchange in the optimization community. Many
Foundations of a Modelica library for general-purpose solvers for solving nonlinear programming (NLP)
chemical process modeling have been built. Multi- problems are interfaced with the AMPL environment.
ple steady-states in ideal two-product distillation were We are aiming to create a ‘Modelica to AMPL’ con-
computed as a proof of concept. The Modelica source verter. One could use the Modelica toolchain to create
code is available at the project homepage. The issues the models conveniently on a GUI. After exporting the
encountered during modeling may be valuable to the Modelica model in AMPL format, the already existing
Modelica language designers. software environments (solvers with AMPL interface,
Keywords: separation, distillation column, tearing AMPL scripts) can be used. Thus an AMPL export fa-
methods, homotopy continuation, bifurcation cility builds a bridge between Modelica users and the
optimization community. Such an implementation ex-
ists (Å KESSON [3]) but it is no longer supported, and
1 Introduction not publicly available.
The object-oriented component-based modeling Our ultimate goal is the development of structure-
methodology in Modelica (F RITZSON [13]) is well- driven optimization methods for solving nonlinear
suited for chemical processes modeling. Accordingly, programming problems (NLP). The structural infor-
Modelica has received attention in the chemical mation (hierarchical components and the connections
engineering literature (S ANDROCK & DE VAAL between them) can be exploited to solve the underly-
[18]). Creating a component-based framework for ing process model efficiently. For example the process
chemical process modeling is one of the goals of our model of the reactive distillation column in C IRIC &
project. This framework then serves as a common M IAO [8], producing ethylene glycol from ethylene
language between mathematicians and chemical oxide and water, has 70 variables and 70 equations.
engineers. The current chemical engineering literature However, the steady-state process model can be solved
is hardly accessible to mathematicians, partly due to by solving univariate equations only, in a proper elim-
the engineering jargon and unwritten traditions. ination order (BAHAREV & N EUMAIER [5]). In other
We created a prototype Modelica implementa- words, the problem is essentially 1-dimensional. Typ-
tion of basic chemical engineering processes. Cur- ically, chemical process models are essentially low-
rently, only steady-state models are supported. dimensional even if their steady-state model is large-
Once this component library is finished, soft- scale.
ware with a graphical user interface, such as the The structural information is difficult to get from
OpenModelica Connection Editor (OMEdit), can be an AMPL source directly, one would rather try to ex-
used to build chemical process models. The process tract it from the flattened AMPL file instead. In prin-
ciple, one could recover the required structural infor- involved in are the connecting equations and specifi-
mation from the flattened model, at least to some ex- cations on their stream variables.
tent. This means that the flattening step throws away
the structural information first, then one must try to
recover it inside a solver. In contrast, the structural class Source
information is programmatically accessible in JMod- output Stream outlet;
elica (Å KESSON et al. [2]) before flattening, and we end Source;
intend to utilize this. class Sink
input Stream inlet;
end Sink;
2 Component-based modeling of che-
mical processes
Chemical processes are well-suited for component- 2.3 Types of equations
based modeling since they are networks of equip-
ments. In turn, it is natural to model the equipments These equations apply to all atomic units in subsec-
hierarchically, as a composite of smaller components. tion 2.4. Only flows of chemicals are considered. Heat
The smallest subcomponents are called atomic units. flows allowing thermal coupling or multidomain mod-
The atomic units are connected by process streams. els would need an extension.
Material balances: A system of C linear equations,
reflecting the conservation of mass.
2.1 Connector class: process streams
Heat balance: A linear equation reflecting the conser-
A process stream S consisting of C substances has vation of energy.
C + 2 independent variables. It is characterized by the Mechanical equilibrium: The outlets have the same
list of variables pressure as the unit. With the exception of the mixer
S = {S. f , S.p, S.H}, and the pressure changer, the pressure of the unit
equals the pressure of its only inlet.
where S. f is an array of size C. See also Table 1. Thermal equilibrium: The enthalpy of the outlets
variable physical meaning SI unit
corresponds to the temperature of the unit. This rela-
f [i] ≥ 0 molar flow rate of substance i = 1 : C mol/s tion is expressed by nonlinear equations (equation of
p≥0 pressure Pa state). If the temperature is not an internal variable of
H enthalpy flowrate J/s the unit then these nonlinear equations are missing.
Table 1: The C + 2 variables characterizing a process Characterizing equations: These equations charac-
stream. terize how the unit works and cannot be changed.
Connections with other units: These equations de-
The graphical representation of process streams is scribe how the units are connected by equating the cor-
by arrows, as shown in Figure 1. responding variables of the involved streams.
Specifications: These equations make the steady state
S model of the unit well-defined. They usually corre-
spond to closed loop control systems. The form of
Figure 1: The graphical representation of stream S. these equations shows large variation: they can be triv-
ial equations as well as complicated nonlinear equa-
The units are connected by streams. The streams tions.
entering the unit are called inlets, while the streams
leaving it are called outlets. The causal flows reflect
the fact that the chemical process streams are directed,
the material can only flow into the direction specified. 2.4 Atomic units

As the name suggests, these units cannot be decom-

2.2 Sources and sinks
posed further to smaller, connected Modelica compo-
Given their simplicity, the easiest way to describe nents. Atomic units implement the UnitOp interface,
these components is by their implementation, see be- that is all the equations listed in Subsection 2.3 apply.
low. The only equations that sources and sinks can be These units are the followings.
 variable physical meaning SI unit
A B C
F ≥0 total molar flow rate mol/s
x[i] ≥ 0 mole fraction of substance i = 1 : C –
Figure 2: Structural types of the atomic units: (A) ∑ x[i] = 1
p≥0 pressure Pa
heat exchanger, pressure changer, reactor; (B) divider, T ≥0 temperature K
flash; (C) mixer. h specific enthalpy J / mol

Table 2: Traditional choice of variables to characterize

1. Mixer a process stream.
2. Heat exchanger
3. Pressure changer
4. Reactor
at atmospheric pressure. The traditional representation
5. Divider
is shown in Table 2.
6. Flash
There are three problems with this representation.
The mixer has multiple inlets and a single outlet. (1) The temperature is uniquely determined by the
All other atomic units have a single inlet and can have other variables and this relation is nonlinear (equation
either one or two outlets. See Figure 2. Some code of state). (2) The material and heat balance equa-
snippets are shown below. The simplicity of the imple- tions are nonlinear because mole fractions are used
mentation is a consequence of proper decomposition. to describe the stream composition. (3) The process
stream definition involves an equality constraint (the
class Mixer mole fractions must sum up to 1).
extends UnitOp(nInlet=2, nOutlet=1);
end Mixer; The first two issues make linear atomic unit mod-
els nonlinear. In particular, the mixer becomes nonlin-
class PressureChanger ear. (The thermodynamically consistent model of the
extends UnitOp(nInlet=1, nOutlet=1);
mixer is nonlinear. However, it is practically always
redeclare class ChangeInPressure=DeltaP;
end PressureChanger; made linear in the chemical engineering literature by
ignoring the so-called heat of mixing.) The mixer is
class Divider the only atomic unit having multiple inlets. Thus, a
extends UnitOp(nInlet=1, nOutlet=2);
nonlinear mixer has a domino effect: many of the com-
Real zeta;
equation posite units are no longer worth decomposing.
outlet[1].f = zeta*inlet[1].f; The temperature can be safely dropped from the
outlet[1].H = zeta*inlet[1].H;
end Divider;
stream definition. It is uniquely determined by the
other variables and it is never needed outside the units.
If, for some reason, the temperature of a stream is
The Divider has one so-called unit parameter, ζ ,
needed, one can always calculate it by running a flash
its value typically comes from specification.
calculation.
The atomic units or the equipments are not referred
to as components in the chemical engineering litera- At first sight, it looks strange to the engineer to drop
ture. Unlike Modelica, the word “component” refers the temperature from the stream definition. Tradition-
to a particular chemical substance in the process. We ally, the temperature is included in the stream variables
call the smallest Modelica components atomic units (e.g. the EMSO model library, DE P. S OARES & S EC -
and the composite Modelica components composite CHI [9]) as it is easily measured in real life with a ther-
units. mometer. Nevertheless, it can be safely excluded.
To make the balance equations linear we use the mo-
2.5 Notes on the process stream definition lar flowrates of the individual substances and the total
enthalpy flowrate in place of the total molar flow rate,
Traditionally, one uses the total molar flowrate, the the mole fractions and the specific enthalpy. This has
mole fractions of the chemical substances, the pressure the beneficial side-effect that the equality constraint
and the temperature to characterize a process stream. disappears since the mole fractions are not present.
In addition, the specific enthalpy is needed to distin- With these changes to the stream definition given in
guish, for example, between boiling water and satu- Table 2, we arrive at the stream definition presented in
rated steam, as they both have a temperature of 100◦ C Table 1.
2.6 Why not Modelica.Fluid?
The Modelica.Fluid library superficially resembles our
library. However, according to the documentation:
Reactor
“The Modelica.Fluid library provides basic interfaces
and components to model 1-dimensional thermo-fluid
flow in networks of pipes. [. . . ] there is the restric-
tion that only media models are supported that have
T, (p,T), (p,h), (T,X), (p,T,X) or (p,h,X) as independent
variables. [. . . ] (Note, T is temperature, p is pressure,
P P
d is density, h is specific enthalpy, and X is a mass
fraction vector).”
The Modelica.Fluid library does not aim at support- P M H R
ing chemical process models. Chemical process mod-
els are different from flows in networks of pipes. P S
We want to preserve the linearity of the material
and heat balances because it plays an important role
in our novel methods. Since the presence of the tem- Figure 3: The reactor of Yi & Luyben and its abstract
perature, the mass / mole fractions or the specific en- decomposition into atomic units. P: pressure changer,
thalpy would make the balance equations nonlinear, M: mixer, H: heat exchanger, R: reactive flash, S: sink.
none of them should appear in the connector class.
As already discussed in subsection 2.5, only the mo-
lar flow rates of the substances, the pressure and the
enthalpy flowrate together guarantee linearity. Unfor-
tunately, the Modelica.Fluid library does not allow this
choice of the independent variables.

2.7 Hierarchical modeling: composite units

We call the smallest Modelica components atomic
units and the composite Modelica components com-
posite units. Often, atomic units only exist on the
level of abstraction. For example the equipment in
Y I & L UYBEN [20] referred to as reactor cannot be
decomposed further into smaller, functioning pieces.
However, it can be modeled by connecting 7 atomic
units and a sink appropriately. None of these units is a
M F
reactor. See Figure 3.
The set of atomic units listed in Subsection 2.4 was
determined by recursively decomposing a variety of
chemical processes. As a result, this set of atomic Figure 4: Hierarchical decomposition of the vapor-
units is sufficient for general-purpose chemical pro- liquid equilibrium cascade into a cascade of stages,
cess modeling. then the decomposition of a stage into a mixer M and
Figure 4 shows an example of hierarchical decom- a flash unit F.
position. The vapor-liquid equilibrium cascade is a
cascade of stages. A stage is a mixer and a flash unit 2.8 Modelica issues encountered
connected appropriately. In real life, the stages are the
smallest, still functioning pieces. The decomposition The unit models are valid only if the molar flowrates
of the stage into a mixer and a flash unit is an abstrac- are nonnegative. This is due to the internal physi-
tion, as the stage does not have a mixer or a flash unit cal structure of the corresponding unit. The natural
inside. Nevertheless, this decomposition is valid for way to impose these nonnegativity constraints is to
modeling. impose it on the molar flowrates and the pressure of
the stream, that is, in the connector class. Inequal-
ity constraints can be represented within the Modelica I II III
language but only by introducing slack variables and
setting the min/max on these variables accordingly.
This approach is rather inconvenient. The Optimica
Figure 6: Schematic figure of a chemical plant. Input:
language extension (Å KESSON et al. [1]) supports in-
raw materials, output: unwanted materials, products
equalities, it is our preferred way of defining inequality
and byproducts. The steps are (I) preparation, (II) re-
constraints.
action and (III) purification.

the investment is spent on separation operation equip-

ments (P RAUSNITZ et al. [17], p. 2).
Many of the practically relevant equipments used in
separation operations (multistage extraction, absorp-
Figure 5: A stage with an optional connection (dashed tion, desorption, stripping and distillation) are inter-
arrow). nally a cascade. Not surprisingly, their mathematical
model can be solved in a sequential manner.
Another difficulty is that Modelica cannot handle Identifying multiple steady states is critical to
arrays of components that have optional connections. proper design, simulation, control, and operation of
All stages have an optional inlet, see Figure 5. This these equipments. Unfortunately, professional simula-
makes the creation of cascades somewhat awkward as tors return only one solution at a time, without indicat-
missing inlets have to be simulated by dummy streams. ing the possible existence of other solutions. Usually,
The details are difficult to explain in text but easy to only one of the steady-states is desired, the so-called
understand from the source code. The reader is re- high purity branch. The other steady states are unde-
ferred to the source code of the VLEcascade, avail- sirable and potentially harmful as they can lead to un-
able from the project homepage at SUPPLEMENTARY expected behavior, meaning that the equipment may
MATERIAL [19]. respond to perturbation in a counterintuitive way.
Given the importance of separation operations, they
have already been modeled in Modelica by several au-
3 Application: separation operations thors, for example D URO & M ORILLA [11], J OOS
et al. [15] and C HANG et al. [7]. Our implementa-
The Modelica implementation discussed in the previ- tion is based on our Modelica component library for
ous section is tested on a separation operation model. general-purpose chemical process modeling. This dis-
The background of the application is briefly presented. tinguishes our implementation from the previous ones.
Then numerical results are given for the particular
benchmark in subsection 3.2.
3.1 Internal physical structure of distillation
A chemical plant takes raw materials as input and
columns
produces products as output. Roughly speaking, three
steps can be distinguished in a chemical plant: prepa- Distillation columns are used in separation operations.
ration, reaction and purification. See Figure 6. Un- The body of a multistage distillation column is a cas-
wanted chemical substances are separated from the cades of stages. In the cascade, the output of one stage
raw input materials in the first step. The unwanted sub- is the input of its two neighbors and vice versa, see
stances may interfere with the reaction in the second Figure 4. This structural information can be exploited
step. The reaction produces the desired products and to solve the underlying process model efficiently.
byproducts. Usually a significant fraction of the reac- The internal physical structure is reflected in the
tants remain unreacted. These unreacted reactants, the mathematical model of the columns. The equations
products and the waste byproducts are separated in the can be evaluated in a sequential manner after guessing
third step, called the purification step. The unreacted just a few variables at one end of the cascade. The es-
reactants are recycled, that is, they are fed back to the sential dimension of the problem is given by the num-
first step. ber of variables that have to be guessed to start the
Both the first and the third step involves separation stage-by-stage computations. The steady-state model
operations. In a typical chemical plant, 40–80% of of distillation columns are essentially low-dimensional
even if their steady-state model is large-scale.
This approach, reducing the large-scale model to

Product purity [mole fraction]

a low-dimensional one, is called the stage-by-stage 0.999
calculation (L EWIS & M ATHESON [16]). Unfortu-
nately, solving the low-dimensional model is very
difficult if not impossible with this method, as it 0.99
shows an extreme sensitivity to the initial estimates.
Thus, currently only high-dimensional techniques are
in use (D OHERTY et al. [10], 13–33). But a proof-of- 0.9
concept method remedies the numerical difficulties of
the stage-by-stage calculation, see BAHAREV & N EU -
MAIER [5].
96 97 98 99 100 101 102
Reflux mass flowrate [kg/min]

3.2 Example: multiple steady-states in ideal

Figure 7: Bifurcation diagram, multiple steady-states
two-product distillation
in ideal two-product distillation. The infeasible
steady-states are represented by dashed lines.
The Modelica implementation discussed in Section 2
is tested on the distillation column presented in JA -
COBSEN & S KOGESTAD [14]. Its main structure cor- 4 Future work
responds to the linear structure presented in Figure 4,
and detailed in subsection 3.1. 4.1 Recovering structural information
Perhaps the simplest distillation columns are the The structural information (connections of the units)
single feed two-product columns with ideal vapor- can help to solve the underlying process model effi-
liquid equilibrium. Even these columns can have mul- ciently, as already mentioned in the introduction and
tiple stead-states (JACOBSEN & S KOGESTAD [14]). in Subsection 3.1. The core equations of the column
One type of multiplicity can occur when the column in Subsection 3.2 are shown below.
has its input specified on a mass or volume basis (e.g.,
mass reflux and molar boilup). This is of high practi- Modelica source:
cal relevance as industrial columns usually have their
inputs specified in this way. connect(cascade.outVapor, condenser.inlet);
connect(condenser.distillate,distillate.inlet);
The model equations are taken from BAHAREV connect(condenser.reflux, cascade.reflux);
et al. [4]. Specifications are: methanol-propanol feed connect(feed.outlet, cascade.feed);
composition, mass reflux flow rate and vapor molar connect(cascade.boilup, reboiler.vapor);
flow rate of the boilup. Heat balances are included in connect(cascade.outLiquid, reboiler.inLiquid);
connect(reboiler.bulk, bulk.inlet);
the model.
In many studies, one is interested in the dependence AMPL source:
of the characteristics on a design parameter (the bifur-
cation parameter) that can be varied, resulting in bi- M_eq{j in 1..N-1}:
furcation diagrams. In this case, the design parameter sum{k in 1..j} F[k]*z[k] + V[j+1]*y[j+1]
= D*y[1] + (sum{k in 1..j} F[k]+V[j+1]-D)*x[j];
is the reflux flowrate specified on mass basis, and the
observed parameter is the product purity. The bifur- M_tot:
cation diagram is given in Figure 7. The model equa- F[N_F]*z[N_F] = D*y[1] + (F[N_F]-D)*x[N];
tions have five distinct solutions in a certain range of
the reflux flow rate. One of the solutions is infeasi- H_eq{j in 1..N-1}:
sum{k in 1..j} qF[k] + V[j+1]*HV[j+1]
ble in practice because it would result in negative flow = V[1]*(HV[1]-HL[0]) + D*HL[0]
rates. The fact that the Modelica implementation gives + (sum{k in 1..j} F[k]+V[j+1]-D)*HL[j];
the expected steady-states suggests that the implemen-
tation of the involved atomic and composite units is The Modelica code is favorable when it comes to
correct. structural information, since it speaks about connec-
tions as clearly as possible. The JModelica environ- [4] Ali Baharev, Lubomir Kolev, and Endre Rév.
ment (Å KESSON et al. [2]) supports programmatic ac- Computing multiple steady states in homoge-
cess to the connectivity information. JModelica will neous azeotropic and ideal two-product distilla-
play an important role in the further development of tion. AIChE Journal, 57:1485–1495, 2011.
our novel methods.
[5] Ali Baharev and Arnold Neumaier. Steady-state
4.2 Optimization multiplicities in reactive distillation: stage-by-
stage calculation revisited. AIChE J., 2012.
Optimization methods are used in almost all areas of
engineering. Typical problems in chemical engineer- [6] Lorenz T. Biegler. Nonlinear programming: con-
ing arise in process design, process control, model de- cepts, algorithms, and applications to chemical
velopment, process identification and real-time opti- processes. SIAM, 2010.
mization, see B IEGLER [6]. Our ultimate goal is to de-
velop structure-driven optimization methods for solv- [7] Chen Chang, Ding Jianwan, Chen Liping, and
ing nonlinear programming problems (NLP). This re- Wu Yizhong. Media Modeling for Distillation
quires an objective function (e.g. minimize cost, max- Process with Modelica/Mworks. In Proceedings
imize yield or profit) to be included in the model. In- of the 8th International Modelica Conference,
equality constraints often required too. Unfortunately, March 20th-22nd, Technical Univeristy, Dres-
Modelica does not support cost function and inequali- den, Germany, pp. 239–245. Linköping Uni-
ties, only the Optimica language extension (Å KESSON versity Electronic Press, Linköpings universitet,
et al. [1]) does. 2011.

[8] Amy R. Ciric and Peizhi Miao. Steady state mul-

4.3 Dynamic simulation tiplicities in an ethylene glycol reactive distilla-
At the moment, only the steady-state model equations tion column. Ind. Eng. Chem. Res., 33:2738–
of the units are implemented in Modelica. It is pos- 2748, 1994.
sible to extend the library to support dynamic simula-
[9] R. de P. Soares and A. R. Secchi. EMSO: A new
tion, but it is not easy in practice. Often, the model
environment for modelling, simulation and op-
equations are not accurately known and the dynamic
timisation. In Computer Aided Chemical Engi-
calculations may involve additional pitfalls.
neering, vol. 14, pp. 947–952. Elsevier, 2003.
Acknowledgements. The research was funded by
[10] M. F. Doherty, Z. T. Fidkowski, M. F. Malone,
the Austrian Science Fund (FWF): P23554.
and R. Taylor. Perry’s Chemical Engineers’
Handbook. McGraw-Hill Professional, 8th ed.,
References 2007.

[1] Johan Åkesson, Tove Bergdahl, Magnus Gäfvert, [11] N. Duro and F. Morilla. A Modelling Methodol-
and Hubertus Tummescheit. Modeling and Opti- ogy for Distillation Columns using Dymola and
mization with Modelica and Optimica Using the Simulink. In Applied Simulation and Modelling.
JModelica.org Open Source Platform. In Pro- ACTA Press, 2003.
ceedings of the 7th International Modelica Con-
ference, Como, Italy, 20-22 September 2009, pp. [12] Robert Fourer, David M. Gay, and Brian Wilson
29–38. Linköping University Electronic Press, Kernighan. AMPL: A Modeling Language for
Linköpings universitet, 2009. Mathematical Programming. Brooks/Cole USA,
2003.
[2] Johan Åkesson, Torbjörn Ekman, and Görel
Hedin. Implementation of a Modelica compiler [13] Peter Fritzson. Principles of Object-Oriented
using JastAdd attribute grammars. Science of Modeling and Simulation with Modelica 2.1.
Computer Programming, 75:21–38, 2010. Wiley-IEEE Press, 2004.

[3] Johan Åkesson. Languages and Tools for Op- [14] E.W. Jacobsen and S. Skogestad. Multiple steady
timization of Large-Scale Systems. PhD thesis, states in ideal two-product distillation. AIChE
Regler, nov 2007. Journal, 37:499–511, 1991.
 [15] Andreas Joos, Karin Dietl, and Gerhard Schmitz.
Thermal separation: An approach for a model-
ica library for absorption, adsorption and rec-
tification. In Proceedings of the 7th Interna-
tional Modelica Conference, Como, Italy, 20-22
September 2009, pp. 804–813. Linköping Uni-
versity Electronic Press, Linköpings universitet,
2009.

[16] W. K. Lewis and G. L. Matheson. Studies in dis-

tillation. Industrial and Engineering Chemistry,
24:494–498, 1932.

[17] John M. Prausnitz, Rüdiger N. Lichtenthaler, and

Edmundo Gomes de Azevedo. Molecular Ther-
modynamics of Fluid-Phase Equilibria. Pren-
tice Hall PTR, Upper Saddle River, NJ, third ed.,
1999.

[18] Carl Sandrock and Philip L. de Vaal. Dynamic

simulation of chemical engineering systems us-
ing OpenModelica and CAPE-OPEN. In 19th
European Symposium on Computer Aided Pro-
cess Engineering, vol. 26 of Computer Aided
Chemical Engineering, pp. 859–864. Elsevier,
2009.

[19] Online supplementary material. Also available

at http://www.mat.univie.ac.at/~neum/
ms/AICHEsuppl/.

[20] Chang K. Yi and William L. Luyben. Design

and control of coupled reactor/column systems–
Part 1. A binary coupled reactor/rectifier system.
Computers & Chemical Engineering, 21(1):25–
46, 1996.

View publication stats