Available online at www.sciencedirect.

com

ScienceDirect
IFAC PapersOnLine 55-7 (2022) 673–678
Simulation
Simulation platform
platform of
of an
an industrial
industrial propylene-propane
propylene-propane splitter
splitter integrated
integrated to
to
Simulation
Advancedplatform of
Process an industrial
Control for propylene-propane
Real Time Optimization splitter integrated
experiments to
Simulation
Advancedplatform of
Process an industrial
Control for propylene-propane
Real Time Optimization splitter integrated
experiments to
Advanced
Simulation Process
platform of Control
an for Real
industrial Time Optimization
propylene-propane experiments
splitter
Advanced Process
Erika
Control forCesar
Oliveira-Silva*,
RealdeTime Optimization
Prada*, Daniel Navia**
experiments to
integrated
Advanced Process
Erika Control forCesar
Erika Oliveira-Silva*,
Oliveira-Silva*, Realde
Erika Oliveira-Silva*, Cesar
Time
Cesar de
Optimization
Prada*,
de Prada*,
Navia**experiments
Daniel Navia**
Prada*, Daniel
Daniel Navia**
*
* Department
Department of of Erika Oliveira-Silva*,
Systems
of Systems
Systems Engineering
Engineering and and Cesar
and Automatic
Automaticde Prada*,
Control,Daniel
Control, SchoolNavia**
School of Industrial
of Industrial Engineering,
Industrial Engineering,
Engineering,
*
* Department
DepartmentUniversityof Systems Engineering
ErikaofOliveira-Silva*,
Engineering and Automatic
Cesar
Automatic Control,
de Prada*,
Control, School
Daniel
School of
Navia**
of Industrial Engineering,
University of Valladolid,
Valladolid, Dr.
Dr. Mergelina
Mergelina s/n,
s/n, 47011,
47011, Valladolid,
Valladolid, Spain
Spain
* DepartmentUniversity of Systemsof
University Valladolid,
ofEngineering
Valladolid, and Dr.
Dr. Mergelina
Automatic s/n,
Mergelina s/n, 47011,
Control,
47011, Valladolid,
School Spain
of Industrial
Valladolid, Spain Engineering,
* DepartmentInstitute of Systems
of
University
Institute of ofEngineering
Sustainable
of Sustainable Valladolid,
Sustainable Processes, Processes, and
Dr.
Processes, Dr. Automatic
Dr. Mergelina
Mergelina
Dr. Mergelina Control,
s/n,
Mergelina s/n, s/n,
47011, School
47011,
s/n, 47011, of Industrial
Valladolid,
Valladolid, SpainSpain
Valladolid, Engineering,
Spain
Institute
InstituteUniversity
of of
Sustainable Valladolid,Processes, Dr. Mergelina
Dr. Mergelinas/n, 47011,
s/n, 47011,
47011, Valladolid,
Valladolid,
Valladolid, Spain Spain
Spain
(e-mail:
(e-mail: [email protected],
[email protected], [email protected]
[email protected] )
Institute (e-mail:
(e-mail: [email protected],
of Sustainable Processes, Dr. Mergelina
[email protected], [email protected]
s/n, 47011, Valladolid,
[email protected] ))) Spain
**
** Dpto.
Dpto. Institute (e-mail:
Ingeniería
Ingeniería ofQuímica
Sustainable
Química Processes,
y Ambiental,
[email protected], Dr. Mergelina
Universidad s/n, Federico
Técnica 47011, Valladolid,
[email protected] Santa Spain
)María, Avd. Vicuña
**
** Dpto.
Dpto. Ingeniería
Ingeniería Química
Química
(e-mail: yyy Ambiental,
Ambiental,
Ambiental,
[email protected],
Universidad
Universidad
Universidad Técnica
Técnica
Técnica Federico
Federico
Federico
[email protected]
Santa
Santa María,
Santa )María, Avd.
María, Avd. Vicuña
Avd. Vicuña
Vicuña
Mackenna,
Mackenna, Campus
Campus San
San Joaquín,
Joaquín, Santiago,
Santiago, Chile
Chile (e-mail:
(e-mail: [email protected])
[email protected])
Mackenna,
** Dpto. Ingeniería
Mackenna,Química Campus
Campus ySan San Joaquín,
Ambiental, Santiago,
Joaquín,Universidad
Santiago, ChileChile (e-mail:
Técnica
(e-mail:[email protected])
Santa María, Avd. Vicuña
[email protected])
** Dpto. Ingeniería
Mackenna,Química Campus ySan Ambiental,
Joaquín,Universidad
Santiago, Chile Técnica
(e-mail:Federico Santa María, Avd. Vicuña
[email protected])
Abstract:
Abstract: The
The present
Mackenna,
present paper
Campus
paper shows
shows San the development
Joaquín,
the development Santiago,of
of aaChile
dynamic
dynamic rigorous
(e-mail:
rigorous model of
[email protected])
model of aa propylene-propane
propylene-propane
Abstract:
Abstract: The
The present
present paper
paper shows
shows the development
thecontroller
development of
of aa dynamic
dynamic rigorous
rigorous model
model of of aa propylene-propane
propylene-propane
splitter
splitter connected
connected to
to an
an industrial
industrial MPC
MPC controller via
via OPC-UA.
OPC-UA. The
The model
model includes
includes the
the material
material and
and energy
energy
splitter
Abstract:
splitter connected
The
connected to
present
to an
an industrial
paper shows
industrial MPC
MPC the controller
development
controller via
via OPC-UA.
of a
OPC-UA.dynamicThe
The model
rigorous
model includes
model
includes the
of
the amaterial and
and energy
propylene-propane
material energy
balances,
balances,
Abstract: thermodynamic
thermodynamic
The present equilibrium,
equilibrium,
paper shows the and
and constitutive
constitutive
development of equations.
equations.
a dynamic Some
Some
rigorous of
of the
the PI
PI
model controllers
controllers
of a presented
presented
propylene-propane in
in
balances,
splitter
balances, thermodynamic
connected
thermodynamicto an equilibrium,
industrial
equilibrium, MPC and
controller
and constitutive
via
constitutive equations.
OPC-UA.
equations. The Some
model
Some of
of the PI
includes
the PI controllers
the material
controllers presented
and
presentedenergy in
in
the
the real
splitter plant
realconnected
plant have also
havetoalso
also been
been
an industrial modeled.
modeled. A dynamic
A dynamic
MPC controller dynamic model
model
viamodel
OPC-UA. requires
requires
TheSomefurther
further information
information
modelofincludes as the
as the
the material sizing
the sizing
sizing of the
of the
and energy the
the
the real
balances,
real plant
plant have
thermodynamic
have also been
been modeled.
equilibrium,
modeled. A
and
A constitutive
dynamic model requires
equations.
requires further
further information
the PI
information as
controllers
as the presented
sizing of
of in
the
equipments,
equipments, heat
heat transfer
transfer coefficients,
coefficients, thermodynamics
thermodynamics data
data and
and Someaa good
good initial
initial value
value for
for the
the state
state and
and
balances,
equipments,
the real thermodynamic
plant heat
have transfer
also equilibrium,
beencoefficients,
modeled. and
A constitutive
thermodynamics
dynamic equations.
model data and
requires a good
further ofinformation
the PI value
initial controllers
as for
the presented
the
sizingstate
of andin
the
equipments,
algebraic
algebraic heat transfer
variables.
variables. The
The been coefficients,
validation
validation of theAthermodynamics
of the model
model was
wasmodel data in
performed
performed and
in open
open a good
loop initial
and
loopinformation
and in value for
in closed
closed loop
loop the state and
simulations.
simulations.
the real plant
algebraic
equipments, havetransfer
variables.
heat also
The modeled.
validation
coefficients, of the dynamic
model was
thermodynamics performedrequires
data in
and further
open a loop
good and in
initial closed
value asfor
the the
loop sizing of and
simulations.
state the
algebraic
The
The simulation variables.
simulation Thewere
results
results validation
were compared
compared of the to model
the was performed
the historical
tothermodynamics
historical data
data of in
of the
the open loop As
process.
process. andfuture
As in closed
future work,
work, loopthe
the simulations.
simulation
simulation
equipments,
The
algebraicsimulation heat
variables. transfer
results
The were coefficients,
compared
validation of the to the
model historical data
was performed dataof and
the
in a good
process.
open loop As initial
and futurevalue
in closed work, for
loop the
the state and
simulation
simulations.
The
platform
platform simulation
created
created results
will
will be
be were
used
used compared
to
to study
study to
new
new the historical
algorithms
algorithms of
ofdataRTO
RTO of the
with
with process.
Modifier
Modifier As future
Adaptation
Adaptation work, the simulation
methodology.
methodology.
algebraic
platform
The simulation
platform variables.
created
created will The
results
will be
be validation
used
were
used to
compared
to study
study of the
new model
to the
new was
algorithms
algorithms performed
historicalofofdataRTO
RTO in
of with
withopen loop
Modifier
the process.
Modifier and in closed
AsAdaptation
future work,
Adaptation loop simulations.
methodology.
the simulation
methodology.
Copyright
Keywords:
The
Keywords:
platform ©
simulation 2022
Real-time
created Thebe
results
Real-timewill Authors.
were toThis
Optimization,
usedcompared
Optimization, study an
is new
to the open
superfractionator, access article
historical
superfractionator,
algorithms Aspen
data
Aspen
of RTO thethe
under
ofDMC,
DMC,
with CC BY-NC-ND
MPC
process.
MPC
Modifier As future license
work, the simulation
Keywords: Real-time
(https://creativecommons.org
Keywords: Real-time Optimization, superfractionator,
/licenses/by-nc-nd/4.0/)
Optimization, superfractionator, Aspen
Aspen DMC,
DMC, MPC Adaptation methodology.
MPC
platform created will be used to study new
Keywords: Real-time Optimization, superfractionator, Aspen DMC, MPC algorithms of RTO with Modifier Adaptation methodology.
Keywords: Real-time Optimization, superfractionator, Aspen DMC, MPC
1. INTRODUCTION
1. INTRODUCTION
INTRODUCTION The paper
The paper
paper is is structured
is structured
structured as as follows:
as follows: Section
Section 2222 describes
follows: Section describes
describes the the
the
1.
1. INTRODUCTION The
The paper is structured as follows: Section describes the
propylene-propane
propylene-propane splitter
splitter and
and the
the FPM
FPM created.
created. Section
Section 3
Dynamic
Dynamic models
models aim
aim
1. INTRODUCTION
to
to simulate
simulate the
the behavior
behavior of
of aa process
process
propylene-propane
The paper
propylene-propane
explains is
the structured
data
splitter
as
splitter and the
follows:
and the FPM
reconciliation
FPM
Section
performed.
created.
2 Section
describes
Sectionthe
created.Section 3334
Dynamic
Dynamic models
models aim
aim to
to simulate
1. INTRODUCTION
simulate the behavior
the behavior of a
of aplant process
process explains
The
explainspaperthethe
the
propylene-propane data splitter
is structured
data reconciliation
as and follows:
reconciliation performed.
Section
performed.
thetheFPM Section
2 describes
created. Sectionthe
Section 3444
over
over time time and
time models
and theythey
theyaimare
are toused
used in different
in different
different stages
stages of of
ofofaaaa aplant
plant life explains
summarizes
life propylene-propane some data reconciliation
characteristic of performed.
Aspen DMC Section
controller.
over
Dynamic
over time and
and they are
are used
used in
simulate
in different stages
the behavior
stages of plant life
process
life summarizes
summarizes
explains thesome
some data characteristic
splitter
characteristic and
reconciliation of
the
of the
theFPMAspen
Aspen
performed. DMC
created.
DMC controller.
Section
controller.
Section 3 4
cycle.
cycle. In the
In the
the design
design stage,
stage, for
for example,
example, a dynamic model is summarizes
Finally, sectionsome characteristic
55 presents the of thevalidation
model Aspen DMC that controller.
compares
Dynamic
cycle.
over time
cycle.
useful
In
Infor
models
the
and theyaim
design
design
studies are to
stage,
used
stage,
of
simulate
for
for
operability
example,
in example,
differentand aaa controllability.
the behavior
stages dynamic
dynamic
dynamic model
ofofa aplant process
model
model is explains
is
life
is
In Finally,
Finally,
summarizes
Finally, section
the
section
sectionsome data
5 presents
presents
characteristic
5 presents the
reconciliation
the
the model
model
of
model the validation
performed.
validation
Aspen
validation DMC that
that
that compares
Section
compares
controller.
compares 4
useful
over
cycle.time
useful forthe
for
Infor and studies
they are
studies
design of used
of
stage,operability
in example,
operability
for differentandstages
and controllability.
of a plant
controllability.
a controllability.
dynamic modellife In
In the
is summarizesopen
the open loop
opensection
loopsome and
and closed
closed the
characteristic loop
loopmodel results
results
of with
with the
thevalidation
Aspen the
the
DMC process
process data
data
controller.
useful
operation, studies
dynamic of operability
simulations are and
used to train operators, In the
the open
Finally, loop
loop and
and closed
5 presents
closed loop
loop results
results with
with the process
that compares
process data
data
operation,
cycle.
operation,
useful In the
for dynamic
design
dynamic
studies simulations
stage,
simulations
of for
operability are
example, used
are used
used
and a to train
dynamic
to train
controllability.operators,
model
train operators,
operators, is
In after
after reconciliation
reconciliation and
and section
section 6
6 presents
presents some
some conclusions.
conclusions.
operation,
validate dynamic
safety simulations
procedures, and are
study to
different operations Finally,
after
the open
after section
reconciliation 5
loop and and
reconciliation presents
and
closed the
section
section model
loop66results validation
presents
presents withsome
some that compares
conclusions.
theconclusions.
process data
validatefor
useful
validate
operation,
validate safety
safetystudies
dynamic
safety procedures,
of operability
procedures,
simulations
procedures, and are
and
and study
used
study different
anddifferent
study controllability.
different operations
operations
to train operations
operators, In the open loop and and closed loop results with theconclusions.
process data
conditions
conditions for
for optimization
optimization and
and control.
control. after reconciliation
2. PROPYLENE sectionPROPANE6 presents some
SPLITTER
operation,
conditions
validate safety
conditions dynamic
for
for simulations
optimization
procedures,
optimization and
and are
control.
andcontrol. useddifferent
study to train operations
operators, after reconciliation 2.
2. PROPYLENE
2. PROPYLENE
PROPYLENE and section PROPANE
PROPANE
PROPANE6 presents SPLITTER
SPLITTER
some conclusions.
SPLITTER
validate
conditions
There are safety
different procedures,
for optimization
ways ofandand study different
control. a distillation column
describing operations 2. PROPYLENE PROPANE SPLITTER
There are
There
conditions
There are different
are different
different ways of
ways
for optimization
ways ofand
of describing
control. aaa these
describing
describing distillation column
distillation
distillation column
column 2. PROPYLENE PROPANE SPLITTER
using
using aa dynamic
dynamic model.
model. Generally,
Generally, these models
models use
use
using
There
using a
are
a dynamic
different
dynamic model.
ways
model. of Generally,
describing
Generally, a these
distillation
these models
models columnuse
use
conservation
conservation laws like
laws like
like mass,
mass, energy,
energy, and momentum,
and amomentum,
momentum, and
and time
time
There
conservation
using
conservation are different
a dynamiclaws
laws ways
like mass,
model.
mass, of energy,
describing
energy,
Generally, and
and distillation
momentum,
these models column
and
and time
use
time
dependent
dependent constitutive
constitutive equations
equations that
that define
define the
the relation
relation
using
dependent
conservation
dependent a dynamic
constitutive
laws like
constitutive model.
mass, Generally,
equations
energy,
equations that
and
that these
define
momentum, models
the
define variables and
the relation use
relation
time
between
between
conservation intensive
intensive variables
variables
laws likevariables
mass, and
and that
energy, extensive
extensive
and momentum, variables and (as
(as
time
between
dependentintensive
between intensive
constitutive variables
equations and
and extensive
extensivedefine variables
theequations
variables (as
relation
(as
equations
equations
dependent of
of state
state and
and
constitutive equilibrium
equilibrium
equations equations).
equations).
that define These
These the equations
relation
equations
between
equations of state
state and
intensive
of andvariables
equilibrium
equilibrium equations).
andequations).
extensive These
variables
These equations
equations (as
form
form a differential-algebraic equations (DAE) system.
between
form
equations
form aaa differential-algebraic
differential-algebraic
intensive
of state andvariables
differential-algebraic equilibrium equations
equations
equations (DAE)These
andequations).
extensive
(DAE)
(DAE) system.
variables
system.
system. equations (as
equations
formorder
In oftostate and equilibrium
a differential-algebraic
make further equations).
equations
experiments(DAE)These system.
of Real equations
Time
In order
In
form
In order
order
a to make
to make further
make
differential-algebraic
to furtherequations
further experiments
experiments
experiments(DAE) of
of
system.
of Real Time
Real
Real Time
Time
Optimization
Optimization (RTO)
(RTO) algorithms,
algorithms, a
a mathematical
mathematical model
model able
able to
to
Optimization
In order the
Optimization (RTO)
to (RTO) algorithms,
make algorithms,
further a mathematical
experiments
a(manipulated
mathematical model
ofvariables
Real able
model able
Time to
to
describe
describe
In order the to effects
effects
make of
of inputs
inputs
further (manipulated
experiments of variables
Real and
and
Time
describe
Optimization
describe the
the on effects
(RTO)
effects of inputs
algorithms,
of inputs a(manipulated
mathematical
(manipulated variables
model
variables ableand
and to
disturbances)
disturbances)
Optimization on
(RTO)the
the outputs
outputs
algorithms, over
over a wider
wider ranges
ranges
mathematical of
of operation
operation
model able is
is
to
disturbances)
describe
disturbances) the on the
effects
on the theoutputs
of
outputs over
inputs wider ranges
(manipulated
overpaperwidershows
ranges of operation
variables
ofdevelopment
operation is and is
required.
required. Therefore,
Therefore, the present
present paper shows the
the development
describe
required.
disturbances)
required. the effects
Therefore,
on
Therefore, the theof
outputs inputs
present
theonpresent over (manipulated
paper
wider shows
paper shows ranges the variables
of development
operation
the development and is Figure 1. Control structure of the propylene propane splitter.
of
of aa simulation
simulation based
based on aa first-principle
first-principle model
model (FPM)
(FPM) of
of an
an Figure
Figure 1. 1. Control
Control structure
structure of of the
the propylene
propylene propane
propane splitter.
disturbances)
of a
required.simulation
of a simulation on
Therefore, the
based outputs
the
based on on a
presentover wider
first-principle
paper ranges
shows
a first-principle model model theof operation
(FPM)
development
(FPM) of
of is
an
an Figure 1. Control structure of the propylene propane splitter.
splitter.
actual
actual
required. industrial-scale
industrial-scale
Therefore, the propylene-propane
propylene-propane
present paper shows splitter,
splitter,
the to
to
developmentbe
be used
used
actual
of a
actual industrial-scale
simulation
industrial-scalebased propylene-propane
on a first-principle
propylene-propane splitter,(FPM)
model
splitter, to be
to be of
used
usedan 2.1 Figure
Modeling 1. Control
and structure of the propylene propane splitter.
simulation
as 2.1 Figure
Modeling and simulation
as
of
as
as aaaaasimulation
actual
benchmark
benchmark
benchmark
industrial-scale
benchmark basedfor
foron
for
for
tuning
tuning
tuning
the
the RTO
a first-principle
the
propylene-propane
tuning the
RTO
RTOmodel
RTO
layer
layer(FPM)
layer
splitter,
layer
before
before
before
to
beforebe of
used
its
its 2.1
an
its
its 2.1 Modeling
Modeling and
and simulation
1. Control structure of the propylene propane splitter.
simulation
implementation
implementation in
in the
the real
real process.
process. The
The simulation
simulation is
is The
2.1
The case
Modeling
case study
study and is
is aa simulation
simulation of a real propane-propylene
actual
implementation
implementation
connected
industrial-scale
as a benchmark via
in for
in
OPC-UA thepropylene-propane
the tuning
with
real process.
real process.
Aspenthe DMCRTOThe splitter,
The layer
controller
to
simulation
before
simulation to
be used
mimic
is The
its
is The
2.1 case
case
Modeling study
study and is aa simulation
simulation
issimulation
simulation of
of aaa real
of real
real propane-propylene
propane-propylene
propane-propylene
connected via OPC-UA with Aspen DMC controller to mimic splitter
splitter of
of a
a Fluid
Fluid Catalytic
Catalytic Cracking
Cracking (FCC)
(FCC) of
of the Petronor
as a
connected
implementation
connected
the behavior
benchmark
via
via of
OPC-UA
in
OPC-UA
the
for
the tuning
with
with
industrial real Aspen
Aspenthe
process.
process
RTO
DMC
DMC The
studied.
layer
controller
controller
In
before
simulation
future
to
to mimic
mimic
work,
its
is splitter
The
splittercaseof
of a
study
a Fluid
Fluid is Catalytic
a simulation
Catalytic Cracking
of
Cracking a (FCC)
real
(FCC) of the
the Petronor
Petronor
propane-propylene
of the Petronor
the behavior of the industrial process studied. In future work, refinery
refinery in
in Bilbao,
Bilbao, northern
northern Spain.
Spain. The
The splitter
splitter studied
studied aims
aims to
implementation
the
the
aa RTO
behavior
connected
behavior via
layer
of
of
with
in
the
OPC-UA
the the
industrial
with
industrial
Modifier
real process.
process
Aspen
process
Adaptation DMC The
studied.
studied.
(Marchetti,
In
controller
Insimulation
future
future to work,
mimic
work,
Chachuat
is The
refinery
splitter
refinerycase in
of
in study
aBilbao,
Fluid
Bilbao, is a simulation
northern
Catalytic
northern Spain. of
Cracking
Spain. a
The
The real
splitter
(FCC)
splitter studied
of the
studied aims to
propane-propylene
Petronor
aims to
to
RTO
connected layer
viaofwith
OPC-UA Modifier Adaptation
withAdaptation
Aspen DMC (Marchetti,
controller Chachuat
to mimic produce
produce
splitter ofhigh
high purity
purity
aBilbao,
Fluid propylene
propylene
Catalytic from
from
Cracking a stream
aa(FCC)
stream of of
of propylene,
thepropylene,
Petronor
atheRTO
aand RTObehavior layer
layer with
with the Modifier
industrial
Modifier Adaptation
process (Marchetti,
studied.
(Marchetti, In future Chachuat
work,
Chachuat produce
refinery high
in purity propylene
northern Spain. fromThe stream
splitter of
studiedpropylene,
aims to
and Bonvin,
Bonvin, 2009)
2009) will
will be added
be added
added to the
to the architecture
thestudied.
architecture in order
inChachuat
order to produce
propane
to produce
propane high
and
and purity
aa northernpropylene
small amount from ofa stream of
impurities propylene,
(C2-C4
the
and
a RTO
and
perform
behavior
Bonvin,
Bonvin,layer
the
of
2009)
with
2009)
tuning
theModifier
industrial
will
will be
be
experiments.
addedprocess
to
Adaptation
to the architecture
(Marchetti,
architecture In future in
in work,
order
order to
to refinery
propane
propane
in Bilbao,
and
high
and aa small
purity small
propylene
small
amount
Spain.
amount
amount fromThe of
of
ofa impurities
splitter studied
impurities
stream of
impurities
(C2-C4
aims to
(C2-C4
propylene,
(C2-C4
perform
aperform
RTO the tuning
layer tuning experiments.
with Modifier Adaptation (Marchetti,inChachuat hydrocarbons).
hydrocarbons). The
The splitter
splitter consists
consists of
of aa total
total condenser,
condenser, aa
perform
and Bonvin, the
the tuning
2009) experiments.
will be added
experiments. to the architecture order to produce
hydrocarbons).
propane
hydrocarbons).
high
and andpurity
The
a 135
The
propylene
splitter
small
splitter
fromof
consists
amount a stream
of
of aaimpurities
consistsstages. total
total
of propylene,
condenser,
(C2-C4
condenser, aa
and Bonvin, 2009) will be added to the architecture in order to partial
partial kettle,
kettle,
propanekettle, and andand 135
a 135 smallequilibrium
equilibrium
amount stages.
of The
The splitter
splitter
impurities has
has
(C2-C4 a
a
perform the tuning experiments. partial
hydrocarbons).
partialcontroller
kettle, and The equilibrium
splitter
135 equilibrium consists stages.
of a The
total
stages.concentrationsplitter
condenser,
The splitter in has
hasthe a
a
DMC
DMC controller to
to maintain
maintain the
the propylene
propylene of aconcentration in
perform the tuning experiments. hydrocarbons).
DMC
DMC controller
partial kettle, and
controller
The
to 135 splitter
to maintain
maintain the
equilibriumconsists
the propylene stages.
propylene
total condenser,
concentration
The splitter
concentration hasthe
in
in the
thea
2405-8963 Copyright © 2022 The Authors. This is an open access article under partial
DMC kettle,
thecontroller
CC BY-NC-NDand 135license
to maintain .the propylene
equilibrium stages.concentration
The splitter in hasthea
Peer review under responsibility of International Federation of Automatic Control. DMC controller to maintain the propylene concentration in the
10.1016/j.ifacol.2022.07.521
674 Erika Oliveira-Silva et al. / IFAC PapersOnLine 55-7 (2022) 673–678

distillate and bottoms product within a range (≥ 97.5% molar thermodynamic equilibrium as will be explained next in
on top and ≤ 10% molar in bottoms); controlling the distillate, Thermodynamics section.
steam flow and top pressure (Figure 1). The process also has
controllers to maintain the level in the accumulator vessel and 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛
ℎ𝑛𝑛𝑛𝑛 = ℎ𝑛𝑛𝑛𝑛+1 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛+1 + 𝐻𝐻𝐻𝐻𝑛𝑛𝑛𝑛−1 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛−1 − ℎ𝑛𝑛𝑛𝑛 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 − 𝐻𝐻𝐻𝐻𝑛𝑛𝑛𝑛 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛 (3)
base at a set point, manipulating the reflux and bottoms 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
flowrate respectively. Feed
2.2 Dynamic rigorous model In the feed tray, total, individual material balances are
The rigorous model is a representation of the real process for presented by Equations (4) and (5). 𝑓𝑓𝑓𝑓 is the feed molar flow
use in the experiments with the DMC and RTO. A propylene and 𝑧𝑧𝑧𝑧𝑗𝑗𝑗𝑗 is the feed molar fraction for each component 𝑗𝑗𝑗𝑗. The
propane splitter is a superfractionator, i.e., a distillation subscript 𝑛𝑛𝑛𝑛𝑓𝑓𝑓𝑓 means the feed stage.
column that performs the separation of components with quite
low relative volatility (<1.2) between the components, so the 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
= 𝑓𝑓𝑓𝑓 + 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛+1 + 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 − 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 − 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 (4)
number of equilibrium stages required is very high. 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
The mathematical model considers the following 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
simplification hypothesis: = 𝑓𝑓𝑓𝑓𝑧𝑧𝑧𝑧𝑗𝑗𝑗𝑗 + 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛+1 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛+1 + 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 (5)
− 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 − 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
• Constant pressure drop.
• The feed consists of four components: propylene, The energy balance is presented by equation (6), similar to
propane, isobutane and ethane. equation (3) .
• The condenser allows sub cooling.
• The column is thermally isolated. 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
ℎ𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 = 𝑓𝑓𝑓𝑓ℎ𝑛𝑛𝑛𝑛 + ℎ𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛+1 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛+1 + 𝐻𝐻𝐻𝐻𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1
• Only the top pressure and accumulator vessel level PI 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 (6)
controllers are modeled. The level inside the reboiler is − ℎ𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 − 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝐻𝐻𝐻𝐻𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
considered constant.
Top tray
The nonlinear dynamic model described was developed with In the top tray, heat and mass balances similar to the internal
EcosimPro (EA Int., 2020). The model has 12090 equations trays could be considered (equations (7) to (9)). 𝑛𝑛𝑛𝑛𝑑𝑑𝑑𝑑 refers to the
and it was solved using the IDAS_SPARSE integration solver top tray and 𝑟𝑟𝑟𝑟 is the molar reflux flowrate.
(Hindmarsh et al., 2005).
𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
Internal Trays = 𝑟𝑟𝑟𝑟 + 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 − 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 − 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 (7)
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
Equations (1) and (2) describe the total and individual 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
component material balances for each stage. In Equation (1), = 𝑟𝑟𝑟𝑟𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑟𝑟𝑟𝑟 + 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 − 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 (8)
the variable 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 is the molar liquid holdup on stage 𝑛𝑛𝑛𝑛. 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛+1 − 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
and 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 represent the liquid molar flow (kmol/h) that comes
from the upper stage and goes to the downer stage respectively 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
and, 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛−1 and 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛 are the vapor flow that comes from the stage ℎ𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 = ℎ𝑟𝑟𝑟𝑟 𝑟𝑟𝑟𝑟 + 𝐻𝐻𝐻𝐻𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛−1 − ℎ𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 (9)
below and goes to the upper stage. − 𝐻𝐻𝐻𝐻𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛
𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 Base and reboiler
= 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛+1 + 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛−1 − 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 − 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛 (1)
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
The column base and reboiler are assumed to be tray number
In Equation (2), the term 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 corresponds to the liquid molar 1. Equation (10) corresponds to the total mass balance at the
fraction of the component 𝑗𝑗𝑗𝑗 in the tray 𝑛𝑛𝑛𝑛 and 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 is the vapor base, where 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 is the molar liquid holdup and 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 is the
molar fraction. liquid molar flow. Equation (11) is the total mass balance in
the reboiler, where 𝑏𝑏𝑏𝑏 is the bottom distillate molar flow and 𝑣𝑣𝑣𝑣1
𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 is the vapor flow leaving the reboiler. As the control level in
= 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛+1 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛+1 + 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛−1 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛−1 − 𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 (2) the reboiler is assumed perfect, the equation (12) calculates the
− 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 molar liquid holdup inside the reboiler and the level ℎ𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟
is constant. Equation (13) corresponds to the individual
The dynamics of the specific liquid enthalpy depends mainly component material balances inside the reboiler. Equation (14)
on the change in the compositions, which is assumed faster is the energy balance in the reboiler. The heat dynamics is
than the dynamics of the total mass. Therefore, its effect on the faster compared to the composition dynamics so equation (14)
energy balance can be negligible, avoiding a high index is used to calculate the vapor flow in each tray until the feed
problem associated with the calculation of the temperature tray, as presented in equation (3). Equation (15) calculates the
(equation (3)). In equation (3), ℎ is the enthalpy of the liquid heat generated 𝑄𝑄𝑄𝑄𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟 by the steam flow 𝐹𝐹𝐹𝐹𝑆𝑆𝑆𝑆 with a heat of
flow and 𝐻𝐻𝐻𝐻 is the enthalpy of the vapor flow. Also, notice that vaporization ∆𝐻𝐻𝐻𝐻𝑉𝑉𝑉𝑉𝑏𝑏𝑏𝑏𝑉𝑉𝑉𝑉 in the pressure 𝑃𝑃𝑃𝑃𝑆𝑆𝑆𝑆 , considering a steam
equation (3) is used to calculate the vapor flow in each tray and quality 𝜏𝜏𝜏𝜏𝑆𝑆𝑆𝑆 . A 𝜏𝜏𝜏𝜏𝑆𝑆𝑆𝑆 of 0 indicates 100% liquid (condensate) and
not the temperature. The temperatures are calculated using the a 𝜏𝜏𝜏𝜏𝑆𝑆𝑆𝑆 of 1 indicates 100% steam.
 Erika Oliveira-Silva et al. / IFAC PapersOnLine 55-7 (2022) 673–678 675

𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑉𝑉𝑉𝑉𝑟𝑟𝑟𝑟_𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐
= 𝑚𝑚𝑚𝑚2 − 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 (10) α= (22)
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑉𝑉𝑉𝑉𝑇𝑇𝑇𝑇_𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐
𝑚𝑚𝑚𝑚𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 ∗𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 Finally, equation (23) represents the total heat that the coolant
𝑏𝑏𝑏𝑏 = 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 − 𝑣𝑣𝑣𝑣1 , 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = 𝐾𝐾𝐾𝐾𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 � (11)
𝐴𝐴𝐴𝐴𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝜌𝜌𝜌𝜌𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 extracts from the process flow. 𝐹𝐹𝐹𝐹𝑊𝑊𝑊𝑊 is the coolant flow, 𝐶𝐶𝐶𝐶𝑃𝑃𝑃𝑃 𝑊𝑊𝑊𝑊 is
𝐴𝐴𝐴𝐴𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟 ℎ𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟 𝜌𝜌𝜌𝜌1 the refrigerant capacity heat and 𝑇𝑇𝑇𝑇𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑊𝑊𝑊𝑊 , 𝑇𝑇𝑇𝑇𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑊𝑊𝑊𝑊 are respectively
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟 = (12) the coolant outlet and inlet temperature.
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀1
𝑑𝑑𝑑𝑑𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,1 𝑄𝑄𝑄𝑄𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 + 𝑄𝑄𝑄𝑄𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = 𝐹𝐹𝐹𝐹𝑊𝑊𝑊𝑊 𝐶𝐶𝐶𝐶𝑃𝑃𝑃𝑃 𝑊𝑊𝑊𝑊 �𝑇𝑇𝑇𝑇𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑊𝑊𝑊𝑊 − 𝑇𝑇𝑇𝑇𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑊𝑊𝑊𝑊 � (23)
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟 = 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,2 − 𝑣𝑣𝑣𝑣1 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,1 − 𝑏𝑏𝑏𝑏𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,1 (13)
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 The composition inside the condenser is assumed to be the
ℎ2 𝑚𝑚𝑚𝑚2 − ℎ1 𝑏𝑏𝑏𝑏 + 𝑄𝑄𝑄𝑄𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟 same of the vapor to be condensate.
𝑣𝑣𝑣𝑣1 = (14)
𝐻𝐻𝐻𝐻1 Accumulator vessel
𝑄𝑄𝑄𝑄𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 = 𝜏𝜏𝜏𝜏𝑆𝑆𝑆𝑆 𝐹𝐹𝐹𝐹𝑆𝑆𝑆𝑆 ∙ ∆𝐻𝐻𝐻𝐻𝑉𝑉𝑉𝑉𝑏𝑏𝑏𝑏𝑉𝑉𝑉𝑉 (𝑃𝑃𝑃𝑃𝑆𝑆𝑆𝑆 ) (15) The accumulator is a horizontal vessel where the liquid from
the condenser is collected. One part of the liquid is withdrawn
Condenser as distillate and the rest is returned to the splitter as reflux
The splitter has a flooded condenser to control the top pressure, liquid. The total and partial mass balance can be written as (24)
manipulating the flooded area of the condenser using a control and (25). The accumulator vessel has a huge volume so an
valve beneath. An increase in the condensate flow decreases energy balance is also necessary (26).
the liquid level inside the condenser and the area available for 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚
condensation increases (Luyben, 2017). = 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 − 𝑟𝑟𝑟𝑟 − 𝑑𝑑𝑑𝑑 (24)
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
Equations (16)-(23) describe the condenser with sub-cooling 𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑐𝑐𝑐𝑐
liquid. Equation (16) is the vapor molar material balance inside = 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 𝑦𝑦𝑦𝑦𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 − 𝑟𝑟𝑟𝑟𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑟𝑟𝑟𝑟 − 𝑑𝑑𝑑𝑑𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑐𝑐𝑐𝑐 (25)
the condenser. 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑣𝑣𝑣𝑣_𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 is the vapor holdup inside the 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
condenser, 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 is the vapor leaving the top tray and 𝑚𝑚𝑚𝑚𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 is the 𝑑𝑑𝑑𝑑(𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 ℎ𝑐𝑐𝑐𝑐 )
amount of vapor that condensates to liquid. Equation (17) = 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 ℎ𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 − 𝑟𝑟𝑟𝑟ℎ𝑐𝑐𝑐𝑐 − 𝑑𝑑𝑑𝑑ℎ𝑐𝑐𝑐𝑐 (26)
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
relates the molar liquid holdup inside the condenser
(𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑟𝑟𝑟𝑟_𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 ) and the flow of liquid entering the liquid phase Note that, actually, 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑟𝑟𝑟𝑟 = 𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑐𝑐𝑐𝑐 as the reflux and distillate are
from condensation, 𝑚𝑚𝑚𝑚𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 , and the liquid flow that goes to the the outlet from the accumulator vessel.
accumulator vessel, 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 .
Thermodynamics
𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑣𝑣𝑣𝑣_𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐
= 𝑣𝑣𝑣𝑣𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 − 𝑚𝑚𝑚𝑚𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 (16) The high pressures (>10 bar) inside the splitter required the use
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 of the Peng Robinson equation of state for the equilibrium
𝑑𝑑𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑟𝑟𝑟𝑟_𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 constant, equations (27) and (28), and enthalpies calculations
= 𝑚𝑚𝑚𝑚𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 − 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 (17) (Matsoukas, 2013). 𝜙𝜙𝜙𝜙𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑖𝑖𝑖𝑖 𝑟𝑟𝑟𝑟,𝑛𝑛𝑛𝑛 is the liquid fugacity coefficient for
𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
component 𝑗𝑗𝑗𝑗 in tray 𝑛𝑛𝑛𝑛, similarly 𝜙𝜙𝜙𝜙 𝑣𝑣𝑣𝑣𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 is the vapor fugacity
Equations (18) and (19), respectively, present the total latent
heat lost by the vapor to condensate to saturated liquid and the coefficient and 𝐾𝐾𝐾𝐾𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 is the liquid-vapour equilibrium constant.
total sensible heat lost to cool down the saturated liquid to a 𝑦𝑦𝑦𝑦𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛
temperature below the bubble point. Both latent and sensible 𝐾𝐾𝐾𝐾𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 =
𝑥𝑥𝑥𝑥𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 (27)
heat transfer processes are very fast, so the dynamics are not
considered. 𝜙𝜙𝜙𝜙𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛
𝐾𝐾𝐾𝐾𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛 = (28)
𝑄𝑄𝑄𝑄𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = 𝑚𝑚𝑚𝑚𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 �𝐻𝐻𝐻𝐻𝑘𝑘𝑘𝑘 − ℎ𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 � (18) 𝜙𝜙𝜙𝜙 𝑣𝑣𝑣𝑣𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑗𝑗𝑗𝑗,𝑛𝑛𝑛𝑛

𝑄𝑄𝑄𝑄𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = 𝑚𝑚𝑚𝑚𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 ℎ𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 − 𝑚𝑚𝑚𝑚𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 ℎ𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 (19)

The temperature values in each tray are calculated by using a
convergence method. First, initial values for the temperature,
Equations (20) and (21) describe the change in temperature for 𝐾𝐾𝐾𝐾 and 𝑥𝑥𝑥𝑥 in each tray are guessed (for 3 of the 4 components
the condensate flow and refrigerant flow, considering the presented in the mixture). Then 𝑦𝑦𝑦𝑦𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 for the three components
equipment design data and the condensate level inside the are calculated by equation (27). The last component molar
condenser. The parameter α describes the percentage of the fraction is given by the sum of molar fractions (always equals
area available for condensation, equation (22). to 1). After that, the last K-value and the fugacity coefficients
are also calculated using (27) and Peng-Robinson. Finally, the
𝑄𝑄𝑄𝑄𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = 𝑈𝑈𝑈𝑈𝑣𝑣𝑣𝑣 (1 − α)A𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 �𝑇𝑇𝑇𝑇𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 − 𝑇𝑇𝑇𝑇𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑊𝑊𝑊𝑊 � (20) fugacity coefficient equations are used as residue equations to
verify the initial values given (temperature and 𝐾𝐾𝐾𝐾).
𝑄𝑄𝑄𝑄𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐_𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = 𝑈𝑈𝑈𝑈𝑟𝑟𝑟𝑟 αA𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 �𝑇𝑇𝑇𝑇𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑎𝑎𝑎𝑎𝑚𝑚𝑚𝑚 − 𝑇𝑇𝑇𝑇𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑊𝑊𝑊𝑊 � (21)
The Murphree tray efficiency, 𝜀𝜀𝜀𝜀𝑛𝑛𝑛𝑛 , is applied to calculate the
actual performance of the splitter, Equation (29).
676 Erika Oliveira-Silva et al. / IFAC PapersOnLine 55-7 (2022) 673–678

𝑦𝑦𝑦𝑦𝑛𝑛𝑛𝑛 = 𝜀𝜀𝜀𝜀𝑛𝑛𝑛𝑛 �𝑦𝑦𝑦𝑦𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,𝑛𝑛𝑛𝑛 − 𝑦𝑦𝑦𝑦𝑛𝑛𝑛𝑛−1 � + 𝑦𝑦𝑦𝑦𝑛𝑛𝑛𝑛−1 (29)

3. DATA RECONCILIATION
In order to find any errors, caused by measured devices
Column hydraulic
malfunction, we used data for almost 4 months in 2019 (data
A simple Francis weir equation that relates the liquid holdup per hour) to verify the mass and energy balance equations.
in the tray with the liquid flow leaving the tray is used
First, as only the propylene concentration is measured in all
(Luyben, 1999), Equation (30). The flow depends on the fluid
the inputs and outputs, we consider the presence of only
mechanics of the tray. 𝑚𝑚𝑚𝑚_𝑣𝑣𝑣𝑣𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑟𝑟𝑟𝑟 is the liquid flow rate over weir,
propane and propylene in the mixture. Then all the volumetric
ℎ𝑟𝑟𝑟𝑟𝑜𝑜𝑜𝑜 is the height of liquid over weir and 𝐿𝐿𝐿𝐿𝑜𝑜𝑜𝑜 is the length of
flow measurements are converted to mass flow, and the molar
weir.
fractions to mass fractions. So, the amount of total mass and
propylene mass that flow in and flow out the column in this
𝑚𝑚𝑚𝑚_𝑣𝑣𝑣𝑣𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑛𝑛𝑛𝑛 = 3.33𝐿𝐿𝐿𝐿𝑜𝑜𝑜𝑜 ℎ𝑟𝑟𝑟𝑟𝑜𝑜𝑜𝑜 1.5 (30) period of time could be calculated. We consider that
Pressure profile integrating non-stationary flowrates over a period larger than
the time constant of the control volumes, makes the
The rigorous model considers that the gas accumulates in the accumulation negligible compared to inflow and outflow
condenser as shown in Equation (16). The pressure drop is values. We also assume that the analyzers are more reliable
constant along the column and the maximum pressure drop than the flow meters, so the following optimization problem
between the base and top is approximately 1 bar. Equation (31) could be written as in equation (33), where
calculates the pressure profile in the column except for the top 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐹𝐹𝐹𝐹 , 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐷𝐷𝐷𝐷 , 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐵𝐵𝐵𝐵 refers to the difference between the
tray, which is given by equation (32) (Luyben, 1999). The term measured and the real value of the feed, distillate and bottoms
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 is the holdup of the vapor phase inside condenser and flows respectively.
𝑧𝑧𝑧𝑧 is the compressibility factor calculated from Peng Robinson.
min 𝜀𝜀𝜀𝜀𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 2 + 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶3 𝐻𝐻𝐻𝐻6 2
𝑃𝑃𝑃𝑃𝑛𝑛𝑛𝑛 = 𝑃𝑃𝑃𝑃𝑛𝑛𝑛𝑛+1 + ∆𝑝𝑝𝑝𝑝 (31) 𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐹𝐹𝐹𝐹 ,𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐷𝐷𝐷𝐷 ,𝑏𝑏𝑏𝑏𝑟𝑟𝑟𝑟𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐵𝐵𝐵𝐵

𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑣𝑣𝑣𝑣_𝑐𝑐𝑐𝑐𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑐𝑐𝑐𝑐 𝑅𝑅𝑅𝑅𝑇𝑇𝑇𝑇𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑛𝑛𝑛𝑛

𝜀𝜀𝜀𝜀𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 = ∑𝑛𝑛𝑛𝑛=0 𝑓𝑓𝑓𝑓
[(𝐹𝐹𝐹𝐹𝑛𝑛𝑛𝑛 + 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐹𝐹𝐹𝐹 ) − (𝐷𝐷𝐷𝐷𝑛𝑛𝑛𝑛 + 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐷𝐷𝐷𝐷 ) −
𝑃𝑃𝑃𝑃𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 = 𝑧𝑧𝑧𝑧 (32)
𝑉𝑉𝑉𝑉𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 (𝐵𝐵𝐵𝐵𝑛𝑛𝑛𝑛 + 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐵𝐵𝐵𝐵 )] (33)
𝑛𝑛𝑛𝑛𝑓𝑓𝑓𝑓
PI controllers 𝜀𝜀𝜀𝜀𝐶𝐶𝐶𝐶3𝐻𝐻𝐻𝐻6 = ∑𝑛𝑛𝑛𝑛=0�(𝐹𝐹𝐹𝐹𝑛𝑛𝑛𝑛 + 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐹𝐹𝐹𝐹 )𝑧𝑧𝑧𝑧𝑟𝑟𝑟𝑟,𝑛𝑛𝑛𝑛 − (𝐷𝐷𝐷𝐷𝑛𝑛𝑛𝑛 +
𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐷𝐷𝐷𝐷 )𝑥𝑥𝑥𝑥𝐷𝐷𝐷𝐷,𝑟𝑟𝑟𝑟,𝑛𝑛𝑛𝑛 − (𝐵𝐵𝐵𝐵𝑛𝑛𝑛𝑛 + 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐵𝐵𝐵𝐵 )𝑥𝑥𝑥𝑥𝐵𝐵𝐵𝐵,𝑟𝑟𝑟𝑟,𝑛𝑛𝑛𝑛 �
The accumulator level and the pressure PI controllers have
s.t. steady state mass balances,
been modelled in the rigorous dynamic model. The first one
process constraints
manipulates reflux flow to maintain the level set point in the
accumulator vessel and the second one manipulates the liquid
flow from the condenser to the accumulator vessel to maintain The problem was solved using the solver in Excel and the
the top pressure in the set point calculated from the DMC. results founded were: 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐹𝐹𝐹𝐹 = 0.1624, 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐷𝐷𝐷𝐷 = 0 and
𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝐵𝐵𝐵𝐵 = 0.0920. Two open loop simulations were performed:
Parameters of the model
the first one using the input flow data and a second one with
A dynamic simulation requires some information of the the input flow bias corrected data. The results of the percentage
equipment involved in the process. The size and overall heat difference for the propylene molar fraction concentration in
transfer coefficient for the vapor phase for the condenser, the distillate are presented in Figure 2. The correction made in the
size of the reboiler and the diameter, weir length and height of process data with the bias decreased the difference between the
the column were founded in the respectively equipment simulation and process data. Similar results were obtained for
specification sheet. Another parameters used in the model are other important simulation variables.
the thermodynamic data. The data was obtained from the
Simulis Thermodynamics software (Simulis Thermodynamics,
2021) and Wauquier (1995).
Initialization and convergence of the model
The splitter rigorous model presented forms a differential-
algebraic equations (DAE) system with 12090 equations that
require the initial values of 553 state variables and 549
algebraic variables (1102 variables in total). To find good
initial values, the methodology proposed in Oliveira-Silva and
Prada (2019) was applied. The methodology constructs
simplified models based on the full rigorous model, applying
assumptions that may not be suitable for the process studied,
but decrease the number of initial values required. In each step,
the simplified models approximate to the complete rigorous Figure 2 Percentage difference between simulations with and
model and a subset of good initial values are found. without reconciliation and process data.
 Erika Oliveira-Silva et al. / IFAC PapersOnLine 55-7 (2022) 673–678 677

4. ASPEN DMC CONTROLLER and simulation have similar gains, but there is a bias between
the values. In Figure 4, one can noticed that the dynamic model
The Aspen DMCplus Build and Aspen DMC3 builder are
is able to match the real gain most of the time.
applications to develop and deploy a MPC (Aspen Technology
Inc, 2021). At each sampling time, Aspen DMC executes a
two-layer architecture: the first one, the optimizer, calculates
the future targets to the second layer, the controller. Then, the
controller calculates the future moves to achieve these targets
and apply the first move to the process.
The DMC optimizer calculates the setpoints to the control
variables (CVs) and manipulated variables (MVs), solving the
Linear Programming (LP) problem (34) subject to the
constraints (≥ 97.5% molar on top and ≤ 10% molar in
bottoms of propylene). The 𝐿𝐿𝐿𝐿𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑟𝑟𝑟𝑟 values are the
incremental cost of increasing the 𝑏𝑏𝑏𝑏 − 𝑑𝑑𝑑𝑑ℎ MV by one unit,
while all other MVs remain constant. By definition, a negative
𝐿𝐿𝐿𝐿𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 tries to maximize the value of the MV and a positive Figure 4 Normalized molar fraction of propylene in bottoms, open
𝐿𝐿𝐿𝐿𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 tries to minimize its value. ∆𝑀𝑀𝑀𝑀𝑉𝑉𝑉𝑉𝑟𝑟𝑟𝑟,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 corresponds to the loop.
change in the 𝑏𝑏𝑏𝑏 − 𝑑𝑑𝑑𝑑ℎ MV in steady state. Note that the main
Model validation in closed loop
objective of the LP is to fulfill the process constraints, as the
controller is unconstrained.
The closed loop simulation architecture is presented in the
𝑛𝑛𝑛𝑛 Figure 5. In order to connect the EcosimPro simulation with
min � ∆𝑀𝑀𝑀𝑀𝑉𝑉𝑉𝑉𝑟𝑟𝑟𝑟,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 × 𝐿𝐿𝐿𝐿𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑟𝑟𝑟𝑟 (34) the Aspen DMC, first a deck was generated. A deck allows any
𝑀𝑀𝑀𝑀𝑉𝑉𝑉𝑉 experiment in EcosimPro to run as a standalone black box
𝑟𝑟𝑟𝑟=1
independent from the main program (EA Int., 2020). After
5. MODEL VALIDATION that, the OPC-UA deck was connected to the Aspen DMC
controller using the Aspen Cim-IO interface manager
Model validation in open loop application. The same DMC application used in the process
was changed to connect to the deck, updating the IO Tag in the
In order to validate the model, some of the historical process
Deployment section in DMC3 Builder, Figure 6. So the DMC
bias-corrected data were compared to the model results. To do
used in the simulation and the plant are the same (same model
so, process data from four consecutive days (one data per hour
and configuration). The DMC were monitored using the Aspen
during four days in June 2019) was used as input to the
APC Web Interface. The code in Python connects to the
EcosimPro simulation, including the values of the manipulated
EcosimPro deck by OPC-UA and call the deck commands to
variables: distillate flowrate, steam flowrate to reboiler and the
reset, run and integrate the simulation over time. During the
top pressure. It means the following validation results does not
simulation, the controller reads the data directly from the deck
consider the DMC controller (open loop simulation). Note that
then calculates the MV control moves (distillate flow, steam
the results were normalized due to confidentiality reasons.
flow and pressure) and send this information back to the deck.
Figure 3 and Figure 4 show the behavior of the normalized
molar fraction of propylene in the distillate and bottoms in
time.
Figure 5 Closed loop simulation architecture.

Figure 3 Normalized molar fraction of propylene in distillate, open Figure 6 Connection between deck and DMC.
loop.
The input data used was the same from the open loop
The figures show that the rigorous model and the process have simulation, except the MVs as already mentioned. The DMC
a very similar dynamic behavior. Figure 3 shows that process controller started after 30 minutes of simulation.
678 Erika Oliveira-Silva et al. / IFAC PapersOnLine 55-7 (2022) 673–678

Figure 7 and Figure 8 show the behavior of the concentrations DMC. The economic cost function will consider the profit by
of propylene in distillate and bottom. Note that as the DMC selling propylene and propane minus the cost to produce
requires past input values to make predictions, the first 20 steam. The products prices will depend if the propylene
hours of simulation should be discarded. The simulation needs specification is fulfilled (≥ 97.5% molar on top and ≤ 10%
some time to update these values. After 48 hours of simulation, molar in bottoms), i.e., the price of products out of
the results approach the process data. specification will be drastically reduced.
ACKNOWLEDGMENTS
This work was supported by JCyL under programs CLU 2017-
09 and UIC 233, as well as by project InCO4In, from Spanish
AEI under project PGC2018-099312-B-C31, both with
FEDER funds. The authors thank the European Social Fund
and the “Consejería de Educación de la Junta de Castilla y
León”.
REFERENCES
Aspen Technology Inc (2021) Aspen DMC3. Available at:
https://www.aspentech.com/en/products/msc/aspen-
dmc3 (Accessed: 22 November 2021).
EA Int. (2020) EcosimPro 6.2, User Manual. Available at:
Figure 7 Normalized molar fraction propylene in distillate, https://www.ecosimpro.com/.
simulation in closed loop with Aspen DMC controller.
Hindmarsh, A. C. et al. (2005) ‘SUNDIALS’, ACM
Transactions on Mathematical Software, 31(3), pp. 363–
396. doi: 10.1145/1089014.1089020.
Luyben, W. L. (1999) ‘Process Modeling , Simulation and
Control Chemical Engineers’, Petroleum Refinery
Engineering.
Luyben, W. L. (2017) ‘Dynamic simulation of flooded
condensers’, Chemical Engineering Research and
Design. Institution of Chemical Engineers, 118, pp. 12–
20. doi: 10.1016/j.cherd.2016.12.001.
Marchetti, A., Chachuat, B. and Bonvin, D. (2009) ‘Modifier-
Adaptation Methodology for Real-Time Optimization’,
Industrial & Engineering Chemistry Research, 48(13),
pp. 6022–6033. doi: 10.1021/ie801352x.
Figure 8 Normalized molar fraction propylene in bottoms,
simulation in closed loop with Aspen DMC controller. Marchetti, A. G. et al. (2020) ‘Modifier Adaptation as a
Feedback Control Scheme’, Industrial and Engineering
6. CONCLUSION AND FUTURE WORK Chemistry Research, 59(6), pp. 2261–2274. doi:
In conclusion, the open loop and closed loop validation show 10.1021/acs.iecr.9b04501.
that the model can predict the dynamic behavior of the process. Matsoukas, T. (2013) ‘Fundamentals of Chemical Engineering
The main differences are in the propylene composition in Thermodynamics: With Applications to Chemical
bottoms. However, for control and optimization purposes, the Processes’, p. 690. doi: 10.1111/j.1475-
dynamic response and gains are more important than the 6803.2004.00096.x.
absolute values of process variables. So, the present model is
considered good enough for the purpose of a virtual plant for Oliveira-Silva, E. and de Prada, C. (2019) ‘Methodology to
RTO experiments. achieve convergence in a rigorous dynamic model of a
superfractionator’, in X Congress of Eurosim. Logroño,
As future work, the RTO+MA algorithm will be included in La Rioja, pp. 1–12.
the Python code and the CVs setpoints calculated will update
the DMC external targets. The inclusion of MA modifiers in Simulis Thermodynamics (2021) Prosim. Available at:
the RTO layer permits to match the necessary conditions of https://www.prosim.net/ (Accessed: 22 November
optimality of the process and model. MA does not require an 2021).
accurate model to find the real optimum of the process, since
the modifiers, i.e., process gradients are calculated correctly Wauquier, J.-P. (1995) Petroleum Refining, Volume 1 - Crude
(Marchetti et al., 2020). Because of that, the model used in Oil, Petroleum Products, Process Flowsheets. Technip,
RTO+MA scheme will be the same linear model of the Aspen Institut Français du Pétrole. Technip. Paris.