Beruflich Dokumente
Kultur Dokumente
In this work, the controllability of different distillation arrangements for the separation of ternary
mixtures is compared. At the same time, the energetic efficiency of each one of the arrangements
is analyzed. Specifically, the direct and indirect sequences of columns, the columns with side
sections, and the divided wall column (DWC) are considered. To compare the controllability of
the different arrangements, the preferred control structures are searched. For the DWC,
operating conditions that favor controllability are discussed. Also, the arrangements have been
compared under different design conditions.
1. Introduction the column with a side stripper and that of the column
with a side rectifier. MRI and condition number (CN)
Different distillation arrangements for the separation controllability indexes were used to quantify the com-
of multicomponent mixtures have been compared in the parisons. They found that the DWC was the most
literature. Comparisons have mainly been based on difficult to control. Their study, however, did not analyze
thermodynamic properties and steady-state conditions. all of the potential given by the DWC complexity.
The main objective has been to find out which distilla-
In this work, using the same controllability indexes
tion arrangements are economically and energetically
as those of Hernandez and Jimenez7 and others, a
more favorable. Conclusions indicate that the divided
comparison between different arrangements is done. To
wall column (DWC) and the Petlyuk column are very
compare the best of every arrangement, optimal designs
attractive in terms of energy consumption.1,2 On the
and optimal operating points from an energetic point
other hand, the control of the DWC has been studied
of view are considered and the best control structures
and it has been seen that the DWC, which is very
are searched. Moreover, to exploit the DWC complexity,
attractive in terms of energy consumption, is also
several DWC operating conditions are compared. The
controllable.3 In this work, the controllability of the
direct and indirect sequences of columns, as well as the
DWC is compared to the controllability of other distil-
columns with side sections and the DWC, are all
lation arrangements. The main objective is to know if
included in the comparison.
the DWC is still attractive when controllability is taken
into account. Energy consumption and controllability depend on the
number of trays of distillation arrangements. For this
In the literature, few works have addressed the
reason, comparisons between distillation arrangements
control of complex distillation arrangements for the
have been carried out under two different design condi-
separation of multicomponent mixtures. Doukas and
tions. In section 4, designs are based on the same
Luyben4 studied the control of a side-stream column.
relationship between the reflux ratio (RR) and the
Three years later the same authors studied the control
minimum reflux ratio (MRR), and the different arrange-
of a prefractionator system.5 Alatiqi and Luyben6 were
ments have different numbers of trays. In section 5,
the first to compare the controllability of two different
arrangements with the same numbers of trays are
distillation arrangements for the separation of a ternary
compared.
mixture. The considered arrangements were the direct
sequence of columns and the column with a side
stripper. Controllability was analyzed through the Mo- 2. Case Study
rari resiliency index (MRI) and other frequency-depend-
ent indexes. They concluded that the recycle and A distillation problem has been selected for the
coupling nature of the column with a side stripper comparison study in this work. It consists of the
contributed positively to disturbance attenuation. separation of an equimolar ideal liquid saturated ter-
nary mixture into 0.99 pure products. The relative
The only authors that have included the DWC in a
volatility of the components (A, B, and C, respectively)
comparison between controllability of different distil-
is 4.65:2.15:1, an approximation of a real benzene,
lation arrangements are Hernandez and Jimenez.7 They
toluene, and o-xylene mixture. The control objective is
compared the controllability of the DWC with that of
to meet A, B, and C purity specifications.
Other DWC distillation problems were treated in the
* To whom correspondence should be addressed. Fax: 34
literature by Serra et al.,8 where the separation of
93 4015750. E-mail: maserra@iri.upc.es.
†
Institut de Robòtica i Informàtica Industria, Parc Tecno- different mixtures into 0.99 pure products was analyzed.
lògic de Barcelona. Mixtures with different ease of separation indexes (ESI)
‡
Universitat Politècnica de Catalunya, ETSEIB. Fax: 34 were considered. Authors showed that the best control
93 4017150. E-mail: antonio.espuna@upc.es (A.E.); lpc@ structures do not depend on the mixture. However, the
eq.upc.es (L.P.). controllability indexes do depend on the mixture. It was
10.1021/ie010609o CCC: $25.00 © 2003 American Chemical Society
Published on Web 03/19/2003
1774 Ind. Eng. Chem. Res., Vol. 42, No. 8, 2003
found that the DWC has better controllability for The MRI is the smallest singular value of the open-
mixtures with ESI close to 1. Notice that the ESI of the loop transfer function. It is the poorer gain of the process
mixture considered in this work is ESI ) 1. (poorer sensitivity), which corresponds to specific input
The simple column with a side stream is also de- and output directions. The set of manipulated variables
scribed as one of the attractive options for ternary that gives the largest MRI over the frequency range of
distillation in the literature.9 However, the simple interest is preferred. The CN is the ratio of the maxi-
column with a side stream is not able to solve the mum singular value to the minimum singular value,
considered separation because of the high purity of typically used for the selection of the best set of
products. Therefore, this arrangement is not included manipulated variables. It provides a numerical indica-
in the comparative analysis of this work. tion of the sensitivity balance in a multivariable system.
Large CNs indicate unbalanced sensitivity and also
3. Control Levels, Models, and Controllability sensitivity to changes in process parameters. Therefore,
Indexes sets of manipulated variables with small CNs over the
frequency range of interest are preferred. RGA is a
Because of the large difference between the mass flow matrix that determines the interaction among control
dynamics time constants and the composition dynamics loops in a multivariable process, and it is frequently
time constants, many authors propose to solve the used to select the control pairing. Each element of the
control problem of distillation columns by levels.10 This RGA is defined as the ratio of the open-loop gain for a
is the approach in this work. The lower level is the selected output when all loops in the process are open
stabilization of the process or inventory control, and the to its open-loop gain when all of the other loops are
upper level is the composition control. Therefore, the closed. Pairings that have RGA close to the unity matrix
composition control will be studied over the stabilized at frequencies around the bandwidth are preferred.
arrangements. The closed-loop regulator transfer function may also
As will be explained, pressure is not modeled in this be used to compare the behavior of the controlled
work. Because of that, the inventory control reduces to distillation arrangements.14 It is (I + GM × B)-1GL,
the control of the tank liquid levels (reboilers and where GM is the manipulated variables transfer func-
condensers). In this work, the distillate and bottoms flow tion, B is the feedback controller transfer function, and
rates (external flows) have been chosen as the manipu- GL is the load variables transfer function. The closed-
lated variables used for inventory control. This is the loop regulator transfer function relates the changes in
typical inventory control structure, at least for columns the outputs with the load disturbances when the control
with low to moderate RRs.11 loops are closed. Singular value decomposition of (I +
In agreement with the distillation model for the GM × B)-1GL can be done. Small peaks in the maxi-
control study of a binary simple distillation column mum singular value indicate a low influence of the load
proposed by Skogestad,12 the following set of assump- disturbances even in the worst direction. Therefore,
tions has been considered to model the different distil- small peaks in the maximum singular value are desired.
lation arrangements. These assumptions give quite MRI, CN, RGA, and the closed-loop regulator transfer
simple nonlinear models, which contain the essential function depend on the frequency. For this reason, they
elements driving the control system. The assumptions will be evaluated in the frequency domain. For control
are constant pressure, constant relative volatility of the purposes, it is usually the frequency range correspond-
components, constant molar flows through the column ing to the closed-loop bandwidth that is of main inter-
sections, no vapor holdup, linear liquid dynamics, and est.15 The inverse of the main open-loop time constant
equilibrium in all stages. These assumptions allow the is a characteristic frequency of the system related to the
development of models based on the mass balance and bandwidth. In distillation, the main time constants are
the equilibrium equations of each tray. Energy balance given by the composition dynamics. High-purity col-
equations are avoided thanks to the assumption of umns have very large open-loop time constants. How-
constant vapor molar flows through the column sections ever, the use of feedback changes the dynamics, and the
at steady state. MATLAB13 is used to model the distil- closed-loop time constants may become much shorter.
lation arrangements. The main open-loop time constant will give a first idea
To calculate the controllability indexes, linear models of the range of frequencies of interest because, as a first
are used. These linear models are obtained from lin- approach, it can be considered that the closed-loop time
earization of the nonlinear models described above. A constant (bandwidth) is about 10 times smaller than the
MATLAB13 file is created to obtain the state space linear open-loop time constant. The main time constant of an
models from the nonlinear ones. Small increments in open-loop system can be calculated as the inverse of the
manipulated and load variables around the nominal eigenvalue of the state-space matrix A with smallest
operation and the consequent increments in outputs are magnitude.16 On the other hand, Skogestad12 stated that
used to construct the linear models. Before linearization, the frequency range important for feedback control is
the inventory control loops of the nonlinear models are from about 0.01 to 1 rad/min. This gives another idea
closed. As indicated, the composition control analysis of the range of frequencies of interest.
is done for a specific inventory control, the tuning of
which has been solved previously. 4. Controllability and Energy Consumption
Two of the most important controllability indexes are Comparison
the MRI and CN, which derive from the singular value
decomposition of the transfer function (linear model). In this section, the controllability and energy con-
The relative gain array (RGA) is also typically used to sumption of the different distillation arrangements are
study the controllability of distillation systems. In this compared.
work, these controllability indexes are used to select the 4.1. Chosen Designs. To design the different ar-
best control structures, as well as to compare the rangements, a relationship of RR/MRR ) 1.23 is im-
controllability of the different distillation arrangements. posed. This is a recommended value in industrial
Ind. Eng. Chem. Res., Vol. 42, No. 8, 2003 1775
Figure 5. DWC.
is preferred. The MRIs of the preferred structures of operated at optimal operation and at nonoptimal opera-
both arrangements have very similar values. In Figures tion 3 are shown. There are important differences
8 and 9, the controllability indexes for the column with between the column operated at optimal operation and
a side stripper and the column with a side rectifier are the column operated at operation 3. For the optimal
shown. Comparing the MRI and CN of the columns with operated DWC, at frequencies lower than 0.1 rad/min,
side sections at low frequencies, it can be seen that the the (L, S, V) control structure is clearly the one associ-
preferred structures for the column with a side stripper ated with the best controllability. The main open-loop
(LM, VM, VS/LM, VM, LIQ) have larger MRIs and lower time constant in this case is 0.004 rad/min. Therefore,
CNs than the preferred structures for the column with taking as a first approach that the bandwidth will be
a side rectifier (LM, VM, LR/LM, VM, VAP). Therefore, close to 0.04 rad/min (10 times the main open-loop time
according to the controllability indexes at low frequen- constant), the bandwidth frequency is expected to be
cies, the controllability of the column with a side smaller than 0.1 rad/min. Therefore, L S V is the
stripper is better. However, in an analysis of the expected preferred control structure. Contrarily, the
controllability indexes at higher frequencies, the pre- DWC operated at nonoptimal conditions has (L/V, S,
ferred control structures are others. The main open-loop SPLITD/SPLITB) as the best control structure if the
time constant for the column with a side stripper is bandwidth is lower than 0.1 rad/min, which is the most
0.0045, and for the column with a side rectifier, it is reasonable because the main open-loop time constant
0.0044. It is just at 0.04 rad/min (10 times the main is 0.004 rad/min. Notice that the DWC operated at
open-loop time constant) where the preferred control different conditions; not only does it have different MRI
structures change. If the actual bandwidth is smaller and CN values, but it has different preferred control
than 0.04 rad/min, the column with a side stripper with structures.
(LM, VM, VS) or (LM, VM, LIQ) control structures is In Figures 12 and 13, the MRI and CN values of the
better. best control structures of the indirect sequence, the
In Figures 10 and 11, the MRI and CN of the different column with a side stripper, and the DWC are plotted
control structures of the DWC are plotted. The DWCs together. Comparing the optimal operated arrange-
1778 Ind. Eng. Chem. Res., Vol. 42, No. 8, 2003
Figure 12. MRI of the best control structures for different Figure 14. RGA diagonal elements of the best paired control
arrangements. structures for different arrangements.
Nomenclature
B ) bottoms flow rate
BI ) bottoms flow rate in an upstream column
BII ) bottoms flow rate in a downstream column
BS ) bottoms flow rate in a stripper
D ) distillate flow rate
DI ) distillate in an upstream column
DII ) distillate in a downstream column
DR ) distillate in a rectifier
L ) reflux
LI ) reflux of the upstream column
LII ) reflux of the downstream column
LIQ ) liquid from the stripper to the main column
LM ) reflux of the main column
LR ) reflux in the rectifier
MRR ) minimum reflux ratio
NCB ) last common tray before the wall in a DWC
NCD ) first common tray after the wall in a DWC
NF ) feed tray
Figure 21. CN for inventory control structures with internal
NFI ) feed tray of the upstream column
flows.
NFII ) feed tray of the downstream column
NM ) number of trays of the main column
more competitive then when the price of energy is more NP ) number of trays of the prefractionator
important than the price of column construction. NR ) number of trays of the rectifier
6. Conclusions NS ) tray in the main column connected to the side section
NSTRIP ) number of trays of the stripper
Upon analysis of the MRI, CN, and RGA in the NT ) total number of trays
frequency domain, the controllabilities of the column NTI ) total number of trays of the upstream column
sequences, the columns with side sections, and the DWC NTII ) total number of trays of the downstream column
are compared for a given separation problem. Through RR ) reflux ratio
a constrained optimization, the boilup flow rate of the S ) side-stream flow rate
different distillation arrangements is minimized and the SPLITB ) split in the base of the wall in a DWC
energetic efficiency of the distillation arrangements is SPLITD ) split at the top of the wall in a DWC
also analyzed. V ) boilup
When the direct sequence of columns is compared VI ) boilup of the upstream column
with the indirect sequence of columns and the column VII ) boilup of the downstream column
with a side rectifier is compared with the column with VAP ) vapor from the main column to the rectifier
a side stripper, it is seen that differences in control- VM ) boilup of the main column
lability and boilup flow rates are small. VS ) boilup of the stripper
Taking into account all three controllability indexes xAD ) purity of product A (M)
at the frequency range of interest, it is concluded that xBS ) purity of product B (M)
the indirect column sequence has a better controllability xCB ) purity of product C (M)
than the DWC. When the DWC is compared with the
column with a side stripper, it is not clear which has a Literature Cited
better controllability because the DWC has worse
(1) Petlyuk, F. B.; Platonov, V. M.; Slavinskii, D. M. Thermo-
indexes in a large range of frequencies, but the column dynamically Optimal Method for Separating Multicomponent
with a side stripper has worse indexes just in the range Mixtures. Int. Chem. Eng. 1965, 5 (3), 555-561.
of frequencies where the bandwidth could be. (2) Fidkowski, Z.; Krolikowski, L. Minimum Energy Require-
At optimal operation, the DWC controllability is worse ments of Thermally Coupled Distillation Systems. AIChE J. 1987,
than the controllability of the indirect column sequence. 33 (4), 643-653.
However, operating the DWC at nonoptimal conditions (3) Wolff, E. A.; Skogestad, S. Operation of Integrated Three-
Product (Petlyuk) Distillation Columns. Ind. Eng. Chem. Res.
(not minimized boilup), its controllability is much better, 1995, 34, 2094-2103.
and it is even better than the controllability of the other (4) Doukas, N.; Luyben, W. L. Control of Sidestream Columns
arrangements. Moreover, the energy consumption of the Separating Ternary Mixtures. Inst. Tech. 1978, 25, 43-47.
DWC at these nonoptimal operating conditions is still (5) Doukas, N. P.; Luyben, W. L. Control of an Energy-
lower than the energy consumption of the other distil- Conserving Prefractionator/Sidestream Column Distillation Sys-
lation arrangements. This result is very important tem. Ind. Eng. Chem. Process Des. Dev. 1981, 20, 147-153.
because it indicates that the DWC operated at some (6) Alatiqi, I. M.; Luyben, W. L. Control of a Complex Side-
stream Column/Stripper Distillation Configuration. Ind. Eng.
nonoptimal operating conditions has the best control- Chem. Process Des. Dev. 1986, 25, 762-767.
lability and the lower energy consumption of all of the (7) Hernandez, S.; Jimenez, A. Controllability Analysis of
arrangements. Thermally Coupled Distillation Systems. Ind. Eng. Chem. Res.
Finally, when different comparisons are made be- 1999, 38, 3957-3963.
tween the different distillation arrangements at differ- (8) Serra, M.; Perrier, M.; Espuña, A.; Puigjaner, L. Analysis
ent design conditions, it is seen that the DWC is more of Different Control Possibilities for the Divided Wall Column:
Feedback Diagonal and Dynamic Matrix Control. Comput. Chem.
competitive when it is designed with a large number of Eng. 2001, 25 (4-6), 859-866.
trays. The DWC needs long columns to be really (9) Glinos, K.; Malone, M. F. Optimality regions for Complex
attractive in terms of energy as well as in terms of Column Alternatives in Distillation Systems. Chem. Eng. Res. Des.
controllability. 1988, 66, 229-240.
1782 Ind. Eng. Chem. Res., Vol. 42, No. 8, 2003
(10) Luyben, W. L. Practical Distillation Control; Van Nostrand (16) Skogestad, S.; Postlethwaite, I. Multivariable Feedback
Reinhold: New York, 1992. Control Analysis and Design; Wiley: Chichester, U.K., 1996.
(11) Luyben, W. L.; Tyreus, B. D.; Luyben, M. L. Plantwide (17) HYSYS; Hyprotech Ltd.
Process Control; 1998. (18) Finn, A. J. Consider Thermally Coupled Distillation. Chem.
(12) Skogestad, S. Dynamics and Control of Distillation Col- Eng. Prog. 1993, 41-45.
umns: A Tutorial Introduction. Trans. Inst. Chem. Eng., Part A (19) Serra, M.; Espuña, A.; Puigjaner, L. Improving the Design
1997, 75. of the Divided Wall Column. AIChE Meeting, Los Angeles, 1997.
(13) MATLAB; The MathWorks Inc.: 1998. (20) Serra, M.; Espuña, A.; Puigjaner, L. Control and Optimi-
(14) Luyben, W. L. Process Modeling, Simulation, and Control sation of the Divided Wall Column. Chem. Eng. Proc. 1999, 38,
for Chemical Engineers; McGraw-Hill Publishing Co.: New York, 549-562.
1990. Resubmitted for review October 28, 2002
(15) Skogestad, S.; Jacobsen, E. W.; Morari, M. Inadequacy of Revised manuscript received January 24, 2003
Steady-State Analysis for Feedback Control: Distillate-Bottom Accepted February 10, 2003
Control of Distillation Columns. Ind. Eng. Chem. Res. 1990, 29,
2339-2346. IE010609O