Article

pubs.acs.org/IECR

Real-Time Optimization of an Industrial-Scale Vapor Recompression

Distillation Process. Model Validation and Analysis
Diego. F. Mendoza,† Lina M. Palacio,† José E. A. Graciano,† Carlos A. M. Riascos,†,‡
Ardson S. Vianna, Jr.,† and G. A. Carrillo Le Roux*,†
†
Departamento de Engenharia Química, Escola Politécnica, Universidade de São Paulo, Avenida Prof. Luciano Gualberto, trav. 3, 380,
São Paulo, Brazil
‡
Departamento de Ingeniería Química y Ambiental, Universidad Nacional de Colombia, Avenida Carrera 30, 45-03, Bogotá,
Colombia

ABSTRACT: The modeling, analysis, and simulation validation of an industrial-scale depropenizer column (Refinaria de
Capuava, Mauá, São Paulo) owned by Petrobras S.A. are carried out using equation-oriented and sequential modular approaches.
The model implemented in the equation-oriented environment proved to be suitable for real-time optimization (RTO)
applications due to its robustness, fast convergence, and ability to represent the real process. The analysis of the model allows
better understanding of the process and establishment of boundaries to process specifications and the parameters updated in the
RTO cycle; furthermore, it is shown that feed composition and column pressure drop are critical aspects in the model to
represent the actual process.

1. INTRODUCTION plant disturbances in a fast and robust way to guarantee a new

In the chemical process industry real-time optimization (RTO) optimal point at every RTO cycle. These requirements make
is a crucial tool to increase process profitability in a scenario mechanistic models and equation-oriented (EO) simulation
where feedstock composition and market price variations environments especially suitable for RTO applications.15−17
demand rapid adaptation of the process conditions. The Vapor recompression assisted distillation (VRD) is a heat
benefits derived from the implementation of an RTO strategy integration technique, widely used in the chemical industry to
in a process plant depend on1 the value of increased processing economically separate close-boiling mixtures.18,19 The main
capacity, differences among product prices, specific energy feature of VRD is that the energy of the overhead vapor, with
consumption, number of independent variables and constraints, added external mechanical energy provided by a compressor, is
and the ability to accurately model process responses. To date, employed to boil up the mixture in the reboiler. The pursuit of
there are around 300 RTO applications,2 spanning a wide decreasing energy and capital costs has led to highly integrated
variety of chemical and petrochemical processes.3−9 The pres- VRD designs such as the one implemented at Refinaria de
ent work focuses on generating a model suitable to the RTO of Capuava (Figure 1).
an industrial-scale depropenizer column (Refinaria de Capuava, Although the mathematical modeling of the equipment
Mauá, São Paulo) owned by Petrobras S.A. This is a highly involved in a VRD process is well-known, the highly interlinked
integrated energy-intensive vapor recompression process structure, added to process nonlinearities and the number of
involving close-boiling mixtures, with high reflux ratios and equations used to describe this superfractionator (around
more than 100 stages. 5000) make this simulation particularly difficult to converge,
The main steps of a classical RTO scheme are10,11 steady especially in sequential modular (SM) simulators,20,21 and
state detection, parameter estimation, and optimization. Online partially explains why simplified and pseudo stream approaches
plant data are analyzed in the first step (including data have been previously proposed to simulate and optimize this
reconciliation and gross error detection) to determine, by some process.22,23 These simplifications, together with neglecting the
criterion, if the process is (reasonably) steady. In the second presence of minor components (<1 mol %) in the feed stream
step, the steady state data are used to update some (relevant) (a process disturbance to be taken into account in the RTO
parameters of the model such that it can represent the actual scheme), might affect the outcome of the optimization.
plant conditions as closely as possible. In the third step, the The implementation of a rigorous model of the VRD process
fitted (adjusted) model is used to find a new optimal operation in an EO simulation environment solves the convergence
point, which becomes the new set point for the control system; difficulties generated by the recycle streams,15 which avoids the
then a new RTO cycle starts. previously mentioned simplifications and allows a better
The success of an RTO strategy largely depends on the representation of the process behavior. Other advantages of
mathematical model of the process and its solution. The model
must be flexible to adjust the real behavior within a wide Received: December 4, 2012
interval of process conditions12 and accurate enough to Revised: February 14, 2013
guarantee that the calculated optimum is close to the real Accepted: March 19, 2013
one.13,14 The solution strategy must be able to handle large Published: March 19, 2013

© 2013 American Chemical Society 5735 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

main product at the bottom, stream 3. The overhead product,

mixed with the vapor coming from distribution tank DT,
stream 8, is compressed to increment its condensing temper-
ature in compressor CP. One part of the outlet flow from CP is
sent to condenser HX1, stream 18, while the rest condenses in
reboiler RB, stream 12. A portion of this flow is subcooled in
HX2, stream 13; the remaining part, stream 16, together with
the outlet streams from HX1 and HX2, expands through valves
V3, V2, V4, respectively, returning to DT where a portion of
the liquid is sent to DC as reflux, stream 5, and the other part is
sent to storage, stream 4.

3. MODELING AND SIMULATION ASPECTS

The VRD process was modeled and simulated in the equation-
oriented (EO) environment EMSO and in a sequential
Figure 1. Process flow sheet. modular (SM) process simulator (Aspen Plus version 7.3) to
assess the suitability of these two flow sheeting approaches for
EO with respect to SM environments are17 better computa- RTO applications. This section is devoted to providing details
tional efficiency, increased flexibility in the specification of related to the implementation of these approaches as well as the
degrees of freedom for simulation, a more natural and variables and parameters used to analyze the process.
numerically suitable interface for optimization problems, and 3.1. Equation-Oriented Approach. EMSO27 is an EO
greater autonomy to create and to manipulate the model. The environment able to simulate dynamic and steady state
main drawback of the EO simulation environment is solving the processes. It has a considerable amount of predefined steady
model from poor initial guesses. state and dynamic models for many devices and also allows
EMSO24 (environment for modeling, simulation, and users to easily create their own models due to its simple
optimization) is an EO simulation environment conceived syntax.28
and developed in a joint effort of Brazilian universities and The VRD process comprises a distillation column with 157
Brazilian petrochemical industries to be a tool for educational plates, a total reboiler, one distribution tank, one compressor,
and industrial purposes,24−26 distributed at no cost for two coolers, two mixers, four stream splitters, and four
academic use at http://www.enq.ufrgs.br/alsoc. Some attri- expansion valves. All these devices are available as predefined
butes that make EMSO adequate to implement RTO are a dynamic models; nevertheless, steady state models were created
powerful differential-algebraic solver; the calculation of because the number of variables employed by the predefined
derivatives via automatic differentiation, parameter estimation, dynamic models is far larger than necessary to simulate the
and optimization routines; and its compatibility with other steady state process; this choice considerably reduces the
programs (e.g., MATLAB) and programming languages (e.g., amount of initialization values required to solve the flow sheet.
C, C++, FORTRAN), which allows using external routines The distillation column was modeled using a rigorous model
developed in these languages. including mass transfer aspects via vapor (Murphree)
The comparative analysis of the SM and EO approaches in efficiencies, one per column section. The heat exchange devices
terms of speed and robustness, as well as the capability of the (RB, HX1, and HX2), distribution tank, compressor and
model to represent the real process at Refinaria de Capuava are expansion valves were modeled considering thermodynamic
addressed herein. This contribution is part of the project aspects only, using isentropic efficiency to calculate the actual
“Development and implementation of real-time optimization work demand of the compressor. Further assumptions consist
solutions in an industrial-process unit using equation-oriented in neglecting pressure drop in the distribution tank, heat
approach” that seeks the analysis, development, and imple- exchange devices, splitters, and mixers and heat loss to the
mentation of RTO in an industrial-scale depropenizer column environment. No simplification regarding the number of
(Refinaria de Capuava, Mauá, São Paulo) using an EO process components is adopted; all the components reported in the
simulator developed in Brazil: EMSO. feed stream analysis are considered in this model.
This paper is ordered as follows. A description of the main All the model variables and parameters were properly scaled
features of the actual process is provided in section 2, modeling to obtain a well-posed problem. A summary of the equations
and simulation aspects are addressed in section 3, the used in the implemented model is shown in the Appendix.
comparative analysis of SM and EO simulation approaches, as 3.1.1. Model Specifications. The analysis of degrees of
well as the capability of the model to represent the actual freedom shows that only six variables, together with pressures
process behavior, are presented section 4 and discussed in of all the devices and feed stream, must be specified to solve the
section 5, and the main conclusions of this study are presented model.
in section 6. The specifications chosen are three temperatures (the outlet
temperature of the distillated that condenses in RB, T12, and the
2. PROCESS DESCRIPTION outlet temperatures of heat exchangers HX1, T19, and HX2, T14),
The flow sheet shown in Figure 1 illustrates the main features two flow rates (reflux, F5, and bottom product, F3), and the split
of the VRD process. A low molecular weight hydrocarbon fraction at S4 (defined as the ratio of flows SFS4 = F13/F12). This
mixture expands through valve V1 and enters the distillation choice is motivated by the fact that most of them are part of
column DC, where high-purity propylene is obtained as the the current control scheme adopted in the refinery, as either
overhead product, stream 2, and propane is obtained as the controlled or manipulated variables, and because previous
5736 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

simulations showed that this specification set favors flow sheet Table 1. Process Specifications for the Base Case
convergence.
flow rate (kmol/h) 687.86
3.1.2. Thermodynamic Properties. The equation of state by
temperature (K) 336.25
Peng and Robinson (PR), using classical van der Waals one-
pressure (atm) 25.4
fluid mixing rules, was employed to calculate the phase
stream 1 composition (mol %) 0.02 (ethylene), 64.4 (propylene),
equilibrium and other thermodynamic properties required in 34.77 (propane), 0.34 (isobutene),
the simulation. PR is available in EMSO using the 0.06 (1-butene), 0.03 (cis-2-butene),
0.03 (trans-2-butene), 0.01 (1,3-
thermodynamic complement VRTherm.29 butadiene), 0.30 (isobutane), 0.03
3.1.3. Initialization Strategy. EO simulations require an (n-butane)
initial value (guess) for each variable involved; these values stream 3 flow rate (kmol/h) 261.605
must be good enough to solve the model. The initial values, for stream 5 flow rate (kmol/h) 6081.44
the 5208 variables, of the model programmed in EMSO were splitter S4 ratio F13/F12 (−) 0.66
obtained by simulating a simplified process in Aspen Plus using
PR to calculate thermodynamic properties, including phase
equilibrium. This simplified approach was devised to calculate ΔPDC; and outlet temperatures in condenser HX1, T19, cooler
the equipment independently, not the whole flow sheet, HX2, T14, and reboiler RB, T12.
avoiding convergence problems related to recycle streams.
3.1.4. Equation Solver. EMSO uses the Sundials suite30 to 4. RESULTS
solve large-scale problems involving nonlinear equations, initial- The outcome of the simulations for the base case and for the
value problems in ordinary differential or differential-algebraic model analysis are presented below. Section 4.1 focuses on
equations. Sundials employs an inexact Newton method, called the robustness of the simulation using EO and SM ap-
KINSOL, to solve the nonlinear equation set generated in the proaches, section 4.2 is devoted to the analysis of the model,
model. and section 4.3 shows the capacity of the model to represent
3.2. Sequential Modular Approach. The VRD process, real plant data.
with similar specifications and thermodynamic model, was 4.1. Modeling. 4.1.1. Comparison of EMSO with Aspen
implemented in one of the most commonly used process Plus. The outcomes of the simulations run in EMSO and Aspen
simulators, Aspen Plus, to compare and validate the EO Plus show good agreement for the base case (see Table 2).
simulation. Note that a sequential modular process simulator High purity propylene (99.98 and 99.96 mol %) is obtained in
such as Aspen Plus cannot directly handle the process process stream 4, with overall molar fractional recoveries of
specification related to the outlet temperature in RB, T12, 95.61 and 96.19% propylene in EMSO and Aspen Plus
since it uses the flow rate F11, not the temperature T12, to simulations, respectively. The differences are related to
calculate the flow sheet. A way to conform with the original discrepancies in the phase equilibrium calculated by PR in
specification is to iteratively change the flow rate until the EMSO (by the thermodynamic package VRTherm) and Aspen
desired outlet temperature in RB is obtained by using the design Plus. This fact can be verified by comparing the equilibrium
specif ication routine available in Aspen Plus. This calculation pressures predicted by PR against experimental data for the
system propane/propylene at 300 K; 31 the maximum
scheme generates many nested cycles that complicate the
discrepancies regarding experimental data are 0.8% (VRTherm)
convergence of the flow sheet.
and 0.1% (Aspen Plus). This result suggests that PR parameters
3.3. Base Case. For the base case, an equilibrium distillation
used by EMSO and Aspen Plus are able to adequately predict
column (i.e., Murphree efficiencies, EMV s = EMV
r = 1) at 9.71 atm,
the vapor−liquid equilibrium of the system.
with negligible pressure drop (ΔPDC = 0), a compressor
The largest difference between the simulations is the vapor
working between 9.71 atm (P2) and 16.668 atm (P10), and an flow rate recycled to the compressor, stream 8; this flow rate
isentropic efficiency (ηCP) equal to 0.75 are assumed. Further, has a discrepancy of 1.47% with respect to the flow predicted
the pressure drop is set to 15.69 atm (from 25.4 to 9.71 atm) by Aspen, and can be explained because the vapor fractions in
across valve V1 and 6.958 atm (from 16.668 to 9.71 atm) across expansion valves, which are composition-dependent, are slightly
valves V2, V3, and V4. The temperatures of streams 12, 14, and larger, about 0.8%, in Aspen Plus than in EMSO.
19 are 311.4, 308.6, and 300 K, respectively. The specifications 4.1.2. Initialization and Convergence of the Model. The
for the feed, reflux, and bottom streams as well as the split base case was solved in EMSO using the proposed initialization
fraction at S4 are shown in Table 1. strategy (see section 3.1) without convergence problems. The
3.4. Analysis of the Model. The analysis focuses on three solution obtained in the base case is adopted as the initial guess
aspects: (i) the capability to represent real process data for the simulations required for the model analysis (section 4.2).
(flexibility), (ii) convergence characteristics (speed and robust- This scheme allowed the convergence of the flow sheet for
ness), and (iii) information about key process variables and all cases.
parameters, including upper and lower limits for process The flow sheet, run in a personal computer with Intel Core
specifications and parameters. All this information is useful for i5-2400 processor with 8 GB of RAM, is solved in less than 4 s
the RTO scheme not only because parameter estimation and in all cases. This time is negligible compared with the time
optimization are model dependent tasks, but also because constant of this type of process32 (∼8 h.). This characteristic
convergence must be ensured when process conditions and contributes to fulfilling the requirement of having a new
model parameters change. optimal point for each RTO cycle.
The evaluated process specifications and parameters were the The simulations implemented in Aspen Plus converged for the
isentropic efficiency of the compressor, ηCP; vapor Murphree base case as well as for the majority of process conditions required
efficiencies in stripping, EMV MV
s , and rectification, Er , zones; feed for the model analysis. Aspen Plus presented convergence problems
composition, zF; reflux ratio, RR; total column pressure drop, for large changes in process conditions, mainly in the reflux
5737 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

Table 2. Results Obtained for the Base Case

EMSO Aspen Plus
stream no. F (kmol/h) T (K) propylene (mol %) F (kmol/h) T (K) propylene (mol %)
1 687.86 336.25 64.41 687.86 336.25 64.41
2 6507.69 292.06 99.95 6507.69 291.89 99.96
3 261.6 299.55 0.0647 261.6 299.51 0.0644
4 426.25 291.12 99.98 426.25 291.89 99.99
5 6081.44 291.89 99.98 6081.44 291.89 99.99
6 6158.6 299.53 6.47 6144.91 299.53 6.44
7 6158.6 300.53 6.47 6144.91 300.51 6.44
8 969.79 292.12 99.98 984.30 292.89 99.99
9 7482.41 292.062 99.98 7492.00 291.9 99.99
10 7477.49 323.66 99.95 7492.49 323.47 99.96
11 6626.46 323.65 99.95 6631.79 323.42 99.96
12 6622.19 311.4 99.98 6631.79 311.4 99.99
13 4370.65 311.4 99.98 4376.98 311.4 99.99
14 4370.65 308.6 99.98 4376.98 308.6 99.99
15 4370.65 291.12 99.98 4376.98 291.89 99.99
16 2251.55 311.4 99.98 2254.81 311.4 99.99
17 2251.55 292.04 99.98 2254.81 291.89 99.99
18 855.95 323.65 99.95 860.21 323.47 99.96
19 855.29 300.00 99.98 860.21 300.00 99.99
20 855.29 292.04 99.98 860.21 291.88 99.99
21 7477.49 291.12 99.98 7492.49 291.89 99.99
22 6507.69 292.04 99.98 6507.7 291.89 99.99

stream and the feed composition; the only way to solve the flow
sheet was by repeatedly performing slight changes in the
process variables until the desired specification was obtained.
This strategy worked well in most cases.
Although Aspen Plus does not provide the elapsed time,
the solution time is also negligible compared to the time
needed to stabilize the plant after a perturbation in process
conditions.
4.2. Analysis of the Model. The analysis of the model is
based on a set of simulations performed in EMSO and Aspen
Plus, varying process parameters and specifications. The results
of the simulations in EMSO and Aspen Plus show the same Figure 2. Temperature profiles for different compositions of (A)
tendency for each evaluated parameter and specification; ethane and (B) isobutane in the feed stream.
furthermore, the difference between the values obtained in
the converged flow sheet is similar to that obtained for the base
case (less than 1.5%) up to the point at which Aspen presented 0.999 to 0.960) due to variations in ethane and isobutene,
convergence problems. For this reason, the results shown below respectively.
correspond to the simulations run in EMSO. 4.2.2. Vapor Murphree Efficiencies. Two vapor Murphree
4.2.1. Feed Composition. The influence of minor efficiencies, one for the rectification zone (EMVr ) and one for the
components on the performance of the process is evaluated stripping zone (EMV s ), were defined in order to take into
choosing light (ethane) and heavy (isobutene) components as account mass transfer phenomena in the distillation column.
representatives of this group. The mole fractions of ethane and These efficiencies are varied between 0.1 and 1 in the
isobutene in the feed stream were varied 1 order of magnitude rectification and stripping zones.
regarding the base case, 2 × 10−4−1 × 10−3 (ethane) and 3 × The temperature profiles, obtained for different efficiency
10−3−2.92 × 10−2 (isobutene); this interval was chosen based values (Figure 3), are more affected by changes in the efficiency
on the expected feed composition variations in the VRD of the stripping zone than of the rectification zone. The
process at Refinaria de Capuava. reduction in mass transfer performance in the rectification zone
The temperature profile in the distillation column shows that (low values of EMV
r ) causes an increase of the temperatures of
this process is more sensitive to composition changes of the rectification zone and a decrease of the temperatures of the
isobutene than to variations in ethane (Figure 2). The partial stripping zone. A similar behavior is obtained with respect to
vaporization of the feed stream, F1, across valve V1 causes the efficiencies of the stripping zone, EMV
s . The tendency of the
isobutane to spread to the rectification and stripping zones, temperature profiles at different EMV s values is qualitatively
increasing the column temperature, while changes in ethane similar to that obtained for variations of isobutene in the feed
composition do not affect the temperature profile significantly. stream (cf. Figures 2 and 3). In both cases, the purity of the
In both cases, the composition of the desired product, ethylene, ethylene in stream 2 drops from 99.9 (EMV s = 1) to 96.3 mol %
decreased 1.2 mol % (from 0.999 to 0.987) and 4 mol % (from (EMV
s = 0.4).
5738 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

Figure 5. Influence of pressure drop in the distillation column on (A)

Figure 3. Influence of vapor Murphree efficiencies in the (A) temperature and (B) propylene composition profile.
rectification and (B) stripping zones of the distillation column.
tray 1 and propane in tray 157 changed less than 0.1% when
4.2.3. Reflux Ratio. The influence of reflux ratio, RR (RR = 1 + compared to the simulation with neglected pressure drop.
F5/F4), on product purity, as well as the energy input to the 4.2.5. Efficiency of the Compressor. The influence of the
process, is shown in Figure 4. isentropic compressor efficiency, ηCP, is analyzed by changing it
in the interval 0.3−1. The main changes in the process
associated with this variable are in the recycled vapor, stream 8,
as well as the outlet temperature and energy consumption in
the compressor (Figure 6). The outlet temperature of the

Figure 4. Influence of reflux ratio on (A) the work input to the

compressor, (B) the purity of the propylene in stream 4, and (C) the
ratio between work input and heat transfer in the reboiler.

Figure 6. Influence of compressor efficiency on the (A) flow,

The product purity achieved in the distillation increases (B) outlet temperature, and (C) work in the compressor.
proportionally to the reflux ratio up to a RR around 12.8, little
improvements in purity are achieved for refluxes larger than this compressor decreases as the efficiency increases; these changes
value, and the minimum reflux ratio that meets product are greater for efficiencies below 0.6. A similar tendency is
specifications (mol % of propylene ≥99.5 in stream 2) is shown by energy demand. On the other hand, the outlet flow of
about 12. the compressor tends to increase with the efficiency; this
The energy demand of the compressor increases linearly with increment comes exclusively from the vapor stream recycled to
the reflux ratio and the contribution of the energy supplied to the compressor, stream 8. The changes in outlet temperature,
the process by the compressor decreases asymptotically as the flow, and energy input in the compressor represent 0.01, 0.1,
reflux ratio increases; this energy represents between 12 and and 6.5 from the values obtained for these variables in the base
21% of the heat exchanged in the reboiler. case where ηCP = 0.75.
4.2.4. Pressure Drop in the Distillation Column. The 4.2.6. Bottom Flow Rate. The bottom stream flow rate is
influence of column pressure drop was evaluated in the interval changed in the interval between 230 and 350 kmol/h, where
0−1.25 atm, assuming a linear profile, i.e., constant pressure 261.6 kmol/h is the specification of the base case. Simulations
drop per tray. The pressure drop in the distillation column for flow rates below 230 kmol/h did not converge, while it is
considerably influences the temperature profile of the column possible to increase the bottom flow rate until it matches the
(Figure 5), especially in the stripping zone. Temperature feed flow rate, stream 1. The latter feature is important because
differences between the first tray and the last tray change from the model is able to represent the real situation that takes
7.49 (ΔPDC = 0 atm) to 12.34 K (ΔPDC = 1.25 atm). The heat place when the product is off specification (Figure 7) and the
exchange in the reboiler slightly increases with the pressure distillated product valve is closed.
drop of the column, 0.2% larger for ΔPDC = 1.25 atm than The increase in the bottom flow reduces the heat load in the
when ΔPDC = 0 atm, while the mole fractions of propylene in reboiler and vapor flow in the column as well as the energy
5739 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

temperature of HX2, T14 = 308.6 K. The opposite occurs in the

second case (case 2), in which T19 is kept constant (T19 = 300 K)
and T14 is varied in the interval 291−311 K. The shorter tem-
perature interval in case 2 is due to the heat load of HX2
tending to zero.
In case 1, the heat load HX1 tends to be constant, while HX2
presents a slight decrease in the whole operational interval.
Around 90% of the total heat load is exchanged in HX1 and
10% is exchanged in HX2. In case 2, the heat load distribution
presents larger variations than in case 1: 24−100% of the total
heat load is exchanged in HX1 while 0−76% is exchanged in
HX2. The effect of the outlet temperature in the process is the
regulation of the amount of vapor recycled into the
compressor; lower energy inputs (compression work) corre-
spond to the lowest operational temperatures allowed for each
case. The most beneficial heat-exchange configuration is for an
outlet temperature of 291 K in HX2; this energy requirement
is 7% lower compared to the lowest energy input obtained in
case 1, outlet temperature 281 K in HX1 (Figure 8).
The outlet temperature of the vapor condensed in reboiler,
T12, varies between 301.4 and 321.4 K. This temperature shows
an effect on the energy demanded in the process, similar to the
Figure 7. Bottom flow rate. (A) Influence of bottom flow rate on one of the outlet temperatures of HX1 and HX2, since it
the heat load in the reboiler, QB, and on the energy demand of the regulates the amount of recycled vapor to the compressor. The
compressor, WC. (B) Influence of bottom flow rate on the mole maximum energy recovery, about 8% regarding the base case
fraction of propylene in the final product (stream 4), Zpropy; on the temperature, corresponds to the minimum allowable outlet
fractional recovery of propylene in the distillation column, Propy Rec.; temperature for T12, 301.4 K in this case.
and on the flow rate of the overhead product, F2. 4.3. Comparison with Real Plant Data. A first approach
to assess the fitting capability of the model to the experimental
consumption in the compressor (Figure 7). A flow rate about steady state plant data was performed using efficiency and
250 kmol/h in the bottom stream is the operational point pressure drop parameters (cases 1 and 2 in Table 3), by
that meets process specifications (mol % propylene ≥ 99.5 in minimizing an unweighted least-squares function of temper-
stream 2) with the highest recovery of propylene (96.28 mol %) atures. The temperature profiles of the adjusted model and
and the minimum energy consumption in the compressor measured temperatures show good agreement (Figure 9). The
(3.02 MW). major discrepancy is in the temperature of the last internal tray,
4.2.7. Outlet Temperature in Heat Exchangers and 303.89 K (model) and 305.07 K (experimental).
Reboiler. Two cases were devised to evaluate the influence of The values obtained for the objective function (Table 3)
the outlet temperature specification of the heat exchangers show a significant improvement in the objective function, OF,
(HX1and HX2) on the heat load distribution and work input especially in the stripping zone (from OFs = 3.126 in case 1 to
(Figure 8). In the first case (case 1), the outlet temperature of OFs = 1.386 in case 2), when the fit is done using two (case 2,
HX1, T19, varies from 281 to 340 K keeping constant the outlet ΔPjr = 0.0047, ΔPjs = 0.018 atm) instead of one (case 1, ΔPjr =
ΔPjs = 0.006 atm) pressure drop parameters. The values
obtained for the parameters suggest larger tray pressure drop
and efficiency in the stripping zone than in the rectification
zone.
The influence of efficiency and pressure drop parameters on
model fit was evaluated by generating temperature profiles for
different pressure drops (keeping efficiencies constant, EMV r =
EMs = 1) and for different efficiencies (at constant pressure drop,
ΔPDC = 0 atm). Efficiency parameters (Figure 10) can be
noticed to modify the temperature profile of the middle and
stripping zones of the distillation column (trays 80−157), while
pressure drop parameters (Figure 11) influence the whole
column profile. Furthermore, pressure drop parameters
generate greater temperature changes per tray than efficiency
parameters. Case 3 is addressed in section 5.3.

5. DISCUSSION
5.1. Modeling. Although the reliability and validity of the
Figure 8. Influence of the outlet temperature of heat exhangers on the simulation should be assessed by its capacity to approximate
heat and work loads. (A) Heat load distribution, case 1, T14 constant. real process data, the comparison between EMSO and Aspen
(B) Heat load distribution, case 2, T19 constant. (C) Work input and Plus represents a necessary initial technical validation (model
total heat transferred for cases 1 and 2. correctness test17) of the model developed in the EO
5740 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

Table 3. Influence of Pressure Drop, Efficiency, and Feed Composition on the Model Fit
rectification section stripping section
ΔPr ΔPs
case total per tray EMV
r OFr total per tray EMV
s OFs ΔPDC
1 0.744 0.006 0.95 1.876 0.198 0.006 1.1 3.126 0.942
2 0.583 0.0047 0.3 1.835 0.594 0.018 1 1.3864 1.177
3 0.583 0.0047 0.3 0.664 0.594 0.018 1 0.0079 1.177

environment. The fact that the results obtained by EMSO and

Aspen Plus are similar confirms that the model is adequately
implemented in EMSO.
Results from the current work showed that it is possible to
simulate the process without using the pseudo stream approach
of Choudhari et al.;22 however, the analysis performed in
section 5.2 supports the idea that convergence problems can
arise in the optimization step of this process using an SM
process simulator, which makes the pseudo stream approach a
plausible alternative to mitigate this problem in sequential
modular simulations. On the other hand, the model
implemented in the EO environment proved to be robust
enough to simulate the whole plant for a wide variety of process
conditions without the need to change the flow sheet structure
or the initial value of the variables. The ranges of the process
conditions for which the model was tested, and converged
Figure 9. Adjusted model using pressure drop and Murphree
efficiency parameters.
satisfactorily, are presented in Table 4. This robustness is a

Table 4. Ranges of Convergence for Process Specifications

range of convergence
Specification lower upper
stream 1
ethylene composition (mol %) 0.02 0.2
isobutane composition (mol %) 0.02 0.2
flow rate (kmol/h) 300 2150
stream 3 flow rate (kmol/h) 225 650
stream 5 flow rate (kmol/h) 2750 40000
stream 7 temp (K) 280.11 324.11
stream 14 temp (K) 264.4 331.4
stream 19 temp (K) 260 340
compressor efficiency (−) 0.3 1
EMV
s (−) 0.1 1.5
EMV
r (−) 0.1 1.5
pressure drop (atm) 0 13.26
Figure 10. Influence of efficiency parameters on the model fit.
desirable feature for online optimization, in which the model
must be solved repeatedly in a reasonable time slot, and serves
to confirm why the EO approach is preferable for RTO
applications to the SM one.15
Modern initialization procedures available in commercial EO
simulators, such as gproms, employ models of increasing
complexity to solve the original model.33 Simplified models can
be obtained by linearization,34 by evaluating the model in
limiting conditions, such as total reflux or zero feed,35 or by
neglecting some terms of the original equation set during the
first iterations of the initialization procedure. Homotopy
continuation35 and relaxation34 are mathematical methods
employed by EO simulators to evolve the model complexity to
solve the flow sheet equations from simple (easy to solve)
models. The initial guess can be automatically generated (large-
scale problems), supplied by previous simulation of a similar
problem, or supplied by an experienced engineer with a good
Figure 11. Influence of pressure parameters on the model fit. knowledge of the process.36 These features considerably reduce
5741 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

distillation columns.41−43 The most common practice in

RTO, adopted in the model proposed here, is treating
efficiencies as adjustable parameters updated from steady
state plant data.43,44 Other factors that justify the use of
efficiencies as adjustable parameters in our model are the risk of
convergence difficulties due to the unbounded behavior of EMV
in multicomponent systems,42,45 the uncertainty of available
correlations used to predict efficiencies (±25%), and the fact
that tray efficiencies will be used to represent model
imperfections due to mass transfer, phase equilibrium, and
pressure drop in the column.
The lumped effect of phase equilibrium and pressure drop on
the efficiency parameter does not allow a direct evaluation of
the mass transfer performance based on the numerical values of
EMV; i.e., low values of EMV do not necessarily indicate
inefficient mass transfer, but can be a consequence of the
Figure 12. Influence of feed composition on the adjustment of the
inaccuracies of the pressure and phase equilibrium parameters
model.
used by the model. However, the estimated values of EMV are
expected not to change drastically with slight variations in
the cost and effort (man-hour consumption) involved in the process conditions and, therefore, are expected to be useful to
simulation. assess the overall performance of the column. Murphree
EMSO, in its current state, does not count on automatic efficiencies will be estimated from the steady state column
initialization procedures; therefore, a lot of effort and time is temperature and pressure drop.
invested in finding a good initial guess compared with 5.2.3. Reflux Ratio. Reflux ratio is a key optimization variable
commercial EO and SM process simulators. This current in any distillation process, because it influences the product
drawback can be overcome by exploiting the object-oriented composition (within the limits imposed by phase equilibrium)
structure of the EMSO modeling language that allows the and the energy consumption, both important for the profit-
implementation of general (reusable) initialization procedures, ability of the process. The results show the benefits in energy
similar to those available in commercial simulators. saving using VRD instead of a conventional distillation column.
5.2. Analysis of the Model. 5.2.1. Feed Composition. The The net energy input in the VRD process, i.e., the energy
feed composition analysis focuses on examining the impact supplied by the compressor, is 79−88% lower in comparison to
of the simplification usually adopted for this specific process, a conventional process under the same conditions. This result
namely, disregard the presence of any component different agrees with the energy savings, over 80% when using VRD
from propane and propylene in the feed;22,32 i.e., the multi- instead of conventional distillation, claimed by Annakou and
component mixture is treated as a propane/propylene binary Mizsey.19
system. The results from this work show that this simplification The most favorable reflux is the minimum required for the
must be carefully evaluated because the presence of minor separation task, in this case a reflux ratio of about 12. This value
components might reduce the product purity achieved by the gives an initial idea of the optimal reflux for the real operation,
process, and even generate an off-specification product, which and it is consistent with the operational reflux rates of 14.3 and
agrees with what was discussed by Friedman.37 Additionally, there 13.6 obtained by Choudhari et al.22 and Muhrer et al.32 for a
is a risk that optimum operational conditions obtained assum- propane/propylene VRD, although these refluxes cannot be
ing a propane/propylene binary feed does not hold for the directly compared to the one obtained in the present work
real mixture; furthermore, actual variations in feed composition because process conditions and column size are different.
at Refinaria de Capuava makes the assumption of a binary feed 5.2.4. Pressure Drop. Column pressure drop has consid-
inconvenient. erable effects on the temperature profile of the stripping section
The expected variations in feed flow rate and composition at of the column (Figure 5), which derives from the fact that the
Refinaria de Capuava can be a good reason to implement RTO pressure in tray 1 is the same for all simulations. Hence,
for these processes;1,2 therefore, the multicomponent nature of pressure differences concerning the base case increase down-
the mixture should not be disregarded in a model suitable for ward (from tray 1 to 157) in the distillation column. Note that
RTO, because optimal operational conditions will change as the pressure drop had a negligible effect (less than 0.1%) on the
feed composition varies. composition profile of the desired product (Figure 5).
5.2.2. Murphree Efficiencies. Mass transfer characteristics in Therefore, the main effect of pressure on phase equilibrium is
distillation processes are commonly taken into account using modifying the saturation temperatures and not the equilibrium
vapor Murphree efficiencies, EMV.38 These efficiencies depend compositions.
on flow regime, tray layout, and physical properties of the Column pressure drop depends on column diameter, tray
mixture.38 There are, in general, C − 1 independent EMV per layout, flow rates, and physical properties of the liquid and
tray (where C is the number of components of the mixture), vapor phases.38 Pressure drop can be calculated from available
each one varying from tray to tray. Vapor Murphree efficiencies correlations for sieve trays (e.g., Bennett et al.46); however,
in multicomponent systems can be predicted from tray layout, some parameters of these correlations must be adjusted, in the
mass transfer models, including flow pattern, and correlations parameter estimation step of the RTO cycle, to ensure that
for binary systems using the Maxwell−Stefan approach39,40 or calculated and measured pressure drops are similar. A simpler
treated as a parameter to be estimated in the model. Both approach, adopted herein, is assuming constant pressure drop
approaches have been used for modeling industrial-scale per tray for each column section, using parameters ΔPjs and
5742 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

ΔPjr for stripping and rectification trays, respectively. These imperfect knowledge about phase equilibrium, mass transfer,
parameters will be adjusted from the data measured, namely, flow patterns, and heat losses occurring in the real process.
tray temperatures and total pressure drop. This simplified However, it is worth noting that the parameter values obtained
approach is based on the following assumptions: (i) flow rates make sense with the distillation operation since greater pressure
and physical properties are considerably different for each drop and efficiency is expected for the trays as the liquid flow
column section, (ii) flow rates and physical properties tend to increases, which is the case of the stripping section. The
be similar for each section, and (iii) tray layout and column increase in liquid flow on the trays increases the flow resistance
diameter are the same for the whole column. The simplified to the rising vapor (higher pressure drop) and the contact time
approach reduces the computational burden avoiding the between the liquid and vapor phases (higher efficiencies).
calculation of physical properties; additionally, the simultaneous The experimental temperature reported for the last tray is
adjustment of efficiency (ErMV, EsMV) and pressure drop higher than that predicted by the model; a possible explanation
parameters (ΔPjr , ΔP js ) will alleviate the inaccuracies for this behavior is the presence of unreported (uncharac-
introduced by this simplification. terized) heavy component(s) in the feed. Routine feed
5.2.5. Efficiency of the Compressor. The isentropic composition analysis at Refinaria de Capuava is performed
efficiency of the compressor is an important factor for the for a few components besides propylene and propane; the
process profitability since it determines the energy cost in the compositions of other components in the mixture are inferred
objective (economic) function optimized in the RTO. High from historical data of a more exhaustive (infrequent) analysis.
efficiencies imply better profitability and vice versa. The results A proof of the importance of a good characterization of the feed
obtained demonstrated that the energy demand of the composition is that a slight modification in the mole fraction of
compressor is highly influenced by its isentropic efficiency, one of the minor components (isobutene) notoriously im-
and that an energy-efficient compressor should work above proves the model fit (see Figure 12 and case 3 in Table 3). This
ηCP > 0.7. The isentropic efficiency is a model parameter result suggests that, for maximizing the benefits of the RTO
to be updated in each RTO cycle using the outlet tem- implementation at Refinaria de Capuava, a better character-
perature of the compressor, stream 10. The actual isentropic ization of the feed composition than the one provided by
efficiency, calculated from the steady state plant data available, routine analysis is required.
is close to 0.87, which is within the expected efficiency interval Although the model can properly fit to steady state plant
(ηCP > 0.7). data, there is the need to address the identifiability problem;
An important aspect to keep in mind, regarding efficiency, is i.e., it must be ensured that the model structure and the quality
that compressors are designed to operate efficiently within a and quantity of experimental data available allow the deter-
(narrow) flow interval; this constraint must be taken into mination of the updated parameters because otherwise the
account during the optimization step in the RTO cycle. adjusted model could be unreliable for optimization purposes.
5.2.6. Bottom Flow Rate. The bottom flow rate, stream 3, More elaborate parameter estimation routines than the one
has a function analogous to the reflux ratio in the column used at this stage of the work (unweighted least squares) are
performance; i.e., it determines the distillate purity, the energy required to properly address the estimability problem to
consumption, and the recovery of the desired product. The guarantee a model with good prediction capacity. Studies on
behavior obtained (Figure 7) clearly shows the trade-off practical identifiability have been undertaken in our group and
between energy input and fractional recovery. A profitable show that there are some biased parameter estimation methods
operation is characterized by a trade-off between energy that could properly handle the identifiability problem.
consumption and the recovery of the desired product. The
most favorable bottom flow rate in the simulations is close to
6. CONCLUSIONS AND OUTLOOK
250 kmol/h, a value which cannot be considered as the
optimum flow because it was not obtained from an The highly interlinked vapor recompression distillation process
optimization procedure; it only indicates that there are of an industrial depropenizer was modeled and simulated in
optimum process conditions for this process. sequential modular and equation-oriented environments. The
5.2.7. Outlet Temperature in Heat Exchangers and comparison between these two approaches showed that both
Reboiler. Heat exchangers HX1 and HX2 and reboiler RB are able to simulate the processes in a large range of process
regulate the amount of recycled vapor to the compressor. conditions. It was possible to simulate the process without
Vapor recirculation is necessary because the energy required in using the pseudo stream approach in both environments,
RB is larger than the one that could be supplied by the differently from what a recent publication suggests.22 However,
compression of stream 2 alone and to avoid compressor an equation-oriented environment will be used to execute the
instability (surge). A desirable operational condition is recycling optimization step in future works due its superior character-
the smaller amount of vapor required for a stable operation. istics (speed, robustness, flexibility in process specifications, and
The process conditions used for the base case show that the availability of analytical first order derivatives) as compared to
87.8% of the recycled vapor is used in the reboiler; the rest is the sequential modular approach.
sent to HX1. Better operation conditions can be achieved by The analysis of the model is a valuable tool to gain general
optimizing the heat exchange distribution in RB, HX1, and understanding and to assess the robustness of the model; some
HX2. important conclusions drawn from this analysis are the
5.3. Comparison with Experimental Data. The adjusted following:
temperature profile shows that the model developed is able to (i) Minor components (between 0.001 and 1 mol %) have
fit real plant data using efficiency (EMV MV
r , Es ) and pressure drop significant effects on product composition and temperature
(ΔP jr, ΔPjs ) parameters. These parameters cannot be profile. They should not be disregarded if composition changes
considered the true values for pressure drop or mass transfer in the feed are expected; this also calls attention to the
efficiencies, but only adjusting parameters which lump the importance of good characterization of the feed stream in RTO.
5743 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746
Industrial & Engineering Chemistry Research Article

(ii) Pressure drop is an important feature to take into summation (eq A7), and total flow (eq A8) equations, and
account in the model, mainly because of its effect on phase the thermal and mechanical equilibria of the streams leaving the
equilibrium, especially on saturation (dew and boiling) stage (eqs A9 and A10). F, V, and L denote the molar flow rates
temperatures. of the feed, “V” and “L” represent the internal liquid and vapor
(iii) Parameters EMV r , Es , ΔPjr, ΔPjs, and ηCP can be
MV
streams, x and y are the mole fractions of the liquid and vapor
estimated from the available plant data, to obtain a good phases, hV and hL are the vapor and liquid molar enthalpies, and
representation of the actual process.
Pj and ΔPj are the total pressure and pressure drop in tray j,
(iv) Outlet temperatures in heat exchangers and reboiler as
well as the reflux ratio (or bottom flow rate) are important respectively.
decision variables in the optimization process, because they The expansion valves are modeled as an adiabatic process in
have a direct effect on product composition and recovery as which the outlet stream is in vapor−liquid equilibrium. The
well as on the energy demand of the process. 2C + 5 equations involved in the modeling of the expansion are
The least-squares method used to adjust the model shown in Table 6; these equations come from the component
(parameter estimation) is not suitable to handle gross errors
and the parameter identifiability problem. The next step in this Table 6. Summary of the Equations for Modeling an
project is to implement parameter estimation routines, with a Adiabatic Expansion
robust objective function as the redescending estimator in
FxiF − Vyi − Lxi = 0 (i = 1, ..., C) (A11)
which it is possible to perform data reconciliation, gross error
detection, and parameter estimation in one step.47,48 FhF − Vh − Lh = 0 V L (A12)
The initialization strategy using the results of a simplified (A13)
yi − K ixi = 0 (i = 1, ..., C)
simulation in Aspen Plus (without recycling streams) is
useful to solve the flow sheet in EMSO. However, it is L
T −T =0 V (A14)
desirable to generate an initialization routine independent L
P −P =0 V (A15)
of Aspen or any other process simulator; this can be done C
by exploiting the compatibility of EMSO with programming 1− ∑ yi = 0 (A16)
languages such as C, C++, and FORTRAN. The idea is to i=1
create an application in C++ capable of solving a flow sheet F−V−L=0 (A17)
with simplified thermodynamics that generates a file fully
compatible with the naming convention and structure of
EMSO. and energy balances (eqs A11 and A12), the phase equilibrium

■
condition (eqs A13−A15), the sum of composition, and the
total mole balance.
APPENDIX: MODEL IMPLEMENTED IN THE The equations used to model the total reboiler are shown
EQUATION-ORIENTED ENVIRONMENT in Table 7. The cold side of the total reboiler, identified by
This appendix summarizes the equations used to model the subscript “C”, is the liquid coming from the last tray of the
flow sheet. They correspond to the distillation column, distillation column while the hot side, denoted by subscript
expansion valves, reboiler, and compressor. The adiabatic “H”, is the overheated vapor coming from the compressor.
distillation column was modeled as a collection of individual The cold-side fluid leaves the reboiler as saturated vapor
trays (numbered from top to bottom). The equations used to while the hot-side fluid can exit the reboiler as saturated or
model a tray j of the column (Table 5) comprise 3C + 6 subcooled liquid. Equations A18−A20 correspond to the
equations, viz., the component and energy balances (eqs A1 material balances (total and by component) and the dew-
and A2), efficiency (eqs A3−A5), hydraulic (eq A6), point condition. Equations A21−A31 describe the hot side

Table 5. Summary of the Equations for Modeling an Adiabatic Distillation Tray

FjVyiF V + F jLxiF jL + Vj + 1yij + 1 + Lj − 1xij − 1 − Vjyij + Ljxij = 0 (i = 1, ..., C) (A1)

j

FjVh FVV + F jLh FLL + Vj + 1hjV+ 1 + Lj − 1hjL− 1 − VjhjV + LjhjL = 0 (A2)

j j

EjMV (yij* − yij + 1) − (yij − yij + 1) = 0 (i = 1, ..., C − 1) (A3)

K ijxij − yij* = 0 (i = 1, ..., C) (A4)

C
1− ∑ yij* = 0 (A5)
i=1

Pj − 1 + ΔPj − PjV = 0 (A6)

C
1− ∑ yij = 0 (A7)
i=1

FjV + F jL + Vj + 1 + Lj − 1 − Vj + Lj = 0 (A8)

T jL = TjV (A9)

P jL = PjV (A10)

5744 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746

Industrial & Engineering Chemistry Research

■
Article

Table 7. Summary of the Equations for Modeling a Total REFERENCES

Reboiler (1) White, D. C. Online optimization: What, where and estimating
(A18) ROI. Hydrocarbon Process. 1997, 76, 43.
FCin − FCout = 0
(2) Darby, M. L.; Nikolaou, M.; Jones, J.; Nicholson, D. RTO: An
xiinC − yiout =0 (i = 1, ..., C) (A19) overview and assessment of current practice. J. Process Control 2011,
C
21, 874.
yiout
C
− K iout out
C xiC = 0 (i = 1, ..., C) (A20)
(3) Mercangöz, M.; Doyle, F. J. Real-time optimization of the pulp
(A21) mill benchmark problem. Comput. Chem. Eng. 2008, 32, 789.
TCL,out − TCV,out = 0
(4) Mudt, D. R.; Pederson, C. C.; Jett, M. D.; Karur, S.; McIntyre, B.;
PCL,out − PCV,out = 0 (A22) Robison, P. R.Refinery-wide optimization with rigorous models. In
C Practical Advances in Petroleum Processing; Hsu, C. S., Robinson, P. R.,
1− ∑ xiout
C = 0 (A23) Eds.; Springer Science Business Media: New York, 2006; Vol. 2, p 257.
i=1 (5) Basak, K.; Abhilash, K. S.; Ganguly, S.; Saraf, D. N. On-Line
(A24) Optimization of a Crude Distillation Unit with Constraints on Product
FHin − FHout = 0
Properties. Ind. Eng. Chem. Res. 2002, 41, 1557.
yiinH − xiout (i = 1, ..., C) (A25) (6) Zhijiang, S.; Jinlin, W.; Jixin, Q. Real-time Optimization of
H = 0
Acetaldehyde Production Process. Dev. Chem. Eng. Miner. Process..
yiout
H
− K iout out
H xi H = 0 (i = 1, ..., C) (A26)
2008, 13, 249.
THL,eq − THV,eq =0 (A27) (7) Zanin, A. C.; De Gouvêa, M. T.; Odloak, D. Industrial
implementation of a real-time optimization strategy for maximizing
PHL,out − PHV,out =0 (A28) production of LPG in a FCC unit. Comput. Chem. Eng. 2000, 24, 525.
C (8) Lid, T.; Strand, S. Real-time optimization of a cat cracker unit.
1− ∑ yiout =0 (A29) Comput. Chem. Eng. 1997, 21, S887.
H
i=1 (9) Martin, G. D.; Latour, P. R.; Richard, L. A. Distillation-Closed-
(A30) Loop Optimization of Distillation Energy. Chem. Eng. Prog. 1981, 77,
THout − THL,eq + ΔTHsub = 0
33.
FCinh FL,in + FHinh FV,in − FCouth FV,out − FHouth FL,out =0 (A31) (10) Sequeira, S. E.; Graells, M.; Puigjaner, L. Real-Time Evolution
C H C H
for On-line Optimization of Continuous Processes. Ind. Eng. Chem.
Res. 2002, 41, 1815.
(11) Naysmith, M. R.; Douglas, P. L. Review of Real Time
of the reboiler and include the material balances, the Optimization in the Chemical Process Industries. Dev. Chem. Eng.
boiling-point condition which allows the calculation of the Miner. Process. 2008, 3, 67.
outlet temperature using a subcooling degree, ΔTsub H , and (12) Chachuat, B.; Srinivasan, B.; Bonvin, D. Adaptation strategies
the energy balance around the reboiler. The equation set for real-time optimization. Comput. Chem. Eng. 2009, 33, 1557.
derived for the reboiler can be applied to the other heat- (13) Krishnan, S.; Barton, G. W.; Perkins, J. D. Robust parameter
exchange devices of the flow sheet. estimation in on-line optimizationPart I. Methodology and
The compressor was modeled using isentropic efficiency, simulated case study. Comput. Chem. Eng. 1992, 16, 545.
(14) Forbes, J. F.; Marlin, T. E.; MacGregor, J. F. Model adequacy
η CP, to find the actual outlet temperature of the com- requirements for optimizing plant operations. Comput. Chem. Eng.
pressor; the equations involved in the compressor modeling 1994, 18, 497.
(Table 8) are material balances (eqs A32 and A33), the (15) Meixell, M. D.; Gochenour, B.; Chen, C.-C. Industrial
Applications of Plant-Wide Equation-Oriented Process Modeling
Table 8. Summary of the Equations for Modeling an 2010. In Advances in Chemical Engineering; Sundmache, K., Ed.;
Isentropic Compressor Elsevier: Burlington, MA, 2010; Chapter 3, p 152.
(16) Yip, W. S.; Marlin, T. E. The effect of model fidelity on real-time
F in − F in = 0 (A32) optimization performance. Comput. Chem. Eng. 2004, 28, 267.
yiin − yiout =0 (i = 1, ..., C) (A33) (17) Pantelides, C. C.; Renfro, J. G. The online use of first-principles
models in process operations: Review, current status and future needs.
P out in
− P − ΔPCP = 0 (A34) Comput. Chem. Eng. 2013, 15, 136.
in out (A35) (18) Harwardt, A.; Marquardt, W. Heat-integrated distillation
s −s =0
columns: Vapor recompression or internal heat integration? AIChE
ηCP(h out in
− h ) − (h isen in
−h )=0 (A36) J. 2012, 58, 3740.
(19) Annakou, O.; Mizsey, P. Rigorous investigation of heat pump
assisted distillation. Heat Recovery Syst. CHP 1995, 15, 241.
pressure change through the compressor (eq A34), the (20) Aydin, B.; Benali, M. Simulation and optimization of
isentropic condition (eq A35), and the isentropic efficiency depropanizer, C3-splitter and debutanizer. In Spring Meeting & 5th
expression (eq A36).

■
Global Congress on Process Safety; AIChE: New York, 2009.
(21) Heyen, G.; Ledent, T.; Kalitventzeff, B. A modular package for
AUTHOR INFORMATION simultaneous calculation of complex interlinked separation processes.
Corresponding Author Comput. Chem. Eng. 1994, 18, S69.
*E-mail: galoroux@usp.br. Tel.: 30 91 11 70. Fax: 38 13 23 80. (22) Choudhari, A.; Gune, P.; Divey, J. Distillation optimization by
Notes vapor recompression. Chem. Eng. 2012, 119, 43.
(23) Koninckx, J.On-line Optimization of Chemical Plants Using
The authors declare no competing financial interest.

■
Steady State Models. Ph.D. Dissertation, University of Maryland,
College Park, 1988; p 424.
ACKNOWLEDGMENTS (24) Rodrigues, R.; Soares, R. P.; Secchi, A. R. Teaching chemical
The authors thank Petrobras S.A. for financial support (Project reaction engineering using EMSO simulator. Comput.-Aided Chem.
0050.0068488.11.9). Eng. 2010, 18, 607.

5745 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746

Industrial & Engineering Chemistry Research Article

(25) Valleriote, A.; Dorigo, L.; Secchi, A. R.; Biscaia Jr, E. C. Reduced
Rigorous Models for Efficient Dynamic Simulation and Optimization
of Distillation Columns. In 22nd European Symposium on Computer
Aided Process Engineering; Bogle, I. D. L., Fairweather, M., Eds.;
Elsevier: London, 2012; p 17.
(26) Martins, M. A. F.; Amaro, C. A.; Souza, L. S.; Kalid, R. A.;
Kiperstok, A. New objective function for data reconciliation in water
balance from industrial processes. J. Cleaner Prod. 2010, 18, 1184.
(27) Soares, R. P.; Secchi, A. R. EMSO: A new environment for
modelling, simulation and optimisation. Comput.-Aided Chem. Eng.
2003, 14, 947.
(28) Soares, R. P. User’s and Programming Manual; http://www.enq.
ufrgs.br/alsoc/download/emso/docs/Manual.pdf (accessed Oct 15,
2012).
(29) VRTECH. VRTherm Manual do Usuário. www.vrtech.com.br
(accessed Sept 27, 2012).
(30) Hindmarsh, A. C.; Brown, P. N.; Grant, K. E.; Lee, S. L.; Serban,
R.; Shumaker, D. E.; Woodward, C. S. Sundials: Suite of nonlinear and
differential/algebraic equation solvers. ACM Trans. Math. Software
2005, 31, 363.
(31) Knapp, H.; Doring, R.; Oellrich, L.; Plocker, U.; Prausnitz, J. M.
Vapor-Liquid Equilibria For Mixtures Of Low Boiling Substances;
Behrens, D., Eckermann, R., Eds.; DECHEMA: Frankfurt, 1981.
(32) Muhrer, C. A.; Collura, M. A.; Luyben, W. L. Control of vapor
recompression distillation columns. Ind. Eng. Chem. Res. 1990, 29, 59.
(33) Pantelides, C. C.; Urban, Z. E. Process modeling technology: a
critical review of recent developments. In Proceedings of the 6th
International Conference on Foundations of Computer-Aided Process
Design; CACHE Publications: Austin, TX, 2004; pp 69−82.
(34) Gorczynski, E. W.; Hutchison, H. P. Towards a quasilinear
process simulatorI. fundamental ideas. Comput. Chem. Eng. 1978, 2,
189−196.
(35) Fletcher, R.; Morton, W. Initialising distillation column models.
Comput. Chem. Eng. 2000, 23, 1811−1824.
(36) Pagani, G.; Monforte, A.; d’Armini.; De Mitri, A.. Improving an
equation-oriented package. Comput. Chem. Eng. 1989, 13, 931−945.
(37) Friedman, Y. Z. What’s Wrong With Unit Closed Loop
Optimization. Hydrocarbon Process. 1995, 74, 107−114.
(38) Lockett, M. J. Distillation Tray Fundamentals; Cambridge
University Press: Cambridge, U.K., 1986.
(39) Müller, N. P.; Segura, H. An overall rate-based stage model for
cross flow distillation columns. Chem. Eng. Sci. 2000, 55, 2515.
(40) Chan, H.; Fair, J. R. Prediction of point efficiencies on sieve
trays. 1. Binary systems. Ind. Eng. Chem. Process Des. Dev. 1984, 23,
814.
(41) Klemola, K. T.; Ilme, J. K. Distillation Efficiencies of an
Industrial-Scale i-Butane/n-Butane Fractionator. Ind. Eng. Chem. Res.
1996, 35, 4579.
(42) Rao, D. P.; Goutami, C. V.; Jain, S. A direct method for
incorporation of tray-efficiency matrix in simulation of multi-
component separation processes. Comput. Chem. Eng. 2001, 25, 1141.
(43) Dave, D. J.; Dabhiya, M. Z.; Satyadev, S. V. K.; Ganguly, S.;
Saraf, D. N. Online tuning of a steady state crude distillation unit
model for real time applications. J. Process Control 2003, 13, 267.
(44) Luo, N.; Qian, F.; Ye, Z.-C.; Cheng, H.; Zhong, W.-M.
Estimation of Mass-Transfer Efficiency for Industrial Distillation
Columns. Ind. Eng. Chem. Res. 2012, 51, 3023.
(45) Krishna, R.; Martinez, H. F.; Sreedhar, R.; Standart, G. L.
Murphree point efficiencies in multicomponent systems. Trans. Inst.
Chem. Eng. 1977, 55, 178.
(46) Bennett, D. L.; Agrawal, R.; Cook, P. J. New pressure drop
correlation for sieve tray distillation columns. AIChE J. 1983, 29, 434.
(47) Ö zyurt, D. B.; Pike, R. W. Theory and practice of simultaneous
data reconciliation and gross error detection for chemical processes.
Comput. Chem. Eng. 2004, 28, 381.
(48) Arora, N.; Biegler, L. T. Redescending estimators for data
reconciliation and parameter estimation. Comput. Chem. Eng. 2001, 25,
1585.

5746 dx.doi.org/10.1021/ie303345z | Ind. Eng. Chem. Res. 2013, 52, 5735−5746