Prediction of equilibrium solubility of C[O.sub.2] in aqueous alkanolamines using differential evolution algorithm.
Kundu, M. ; Chitturi, A. ; Bandyopadhyay, S.S. 等
INTRODUCTION
Removal of acid gas impurities, such as C[O.sub.2] and [H.sub.2]S,
from gas mixtures is very important in natural gas processing, hydrogen
purification, treating refinery-off gases, and gas synthesis for ammonia
manufacture. Regenerative chemical absorption of the acid gases, into
solutions of alkanolamines is widely used for gas treating. For the
rational design of gas-treating processes the knowledge of
vapour--liquid equilibrium (VLE) of the acid gas over alkanolamine
solution is required besides the knowledge of mass transfer and
kinetics. The equilibrium solubility determines the minimum circulation
rate of the solvent through the absorber, determines the maximum
allowable concentration of the acid gases in the regenerated solution to
meet the product gas specification, and provides boundary conditions for
solving partial differential equations describing mass transfer coupled
with chemical reactions. The major problem concerning the VLE
measurements of aqueous alkanolamine--acid gas systems, in general, is
that lack of consistency and regularity in the numerous published
values. One problem that is encountered in modelling acid gas treating
is that the experimenters report statistically different partial
pressures of the acid gases at the exactly same conditions (Kundu,
2004). Reliable thermodynamic models can be used confidently to
interpolate or extrapolate experimental data, required for process
design.
In order to have an understanding of those chemically reacting,
multi-component, multiphase systems, a brief discussion about the
chemical equilibria, vapour--liquid phase equilibria, thermodynamic
framework, and activity coefficient model are included, which ensures an
appropriate formulation of the objective function.
The present work requires solving multivariable optimization
problem to determine the interaction parameters of the developed VLE
model. Among the open literature on modelling VLE of C[O.sub.2] in
single and mixed amine solvents, using the traditional, deterministic
techniques, the work of Kent and Eisenberg (1976); Li and Shen (1993);
Deshmukh and Mather (1981); Austgen and Rochelle (1991); Li and Mather
(1994); Kuranov et al. (1996) are the major ones. Most of the
traditional optimization algorithms, like LM, based on gradient methods
have the possibility of getting trapped at a local optimum depending
upon the degree of non-linearity and initial guess. Unfortunately, none
of the traditional algorithms guarantee the global optimal solution, but
genetic algorithms (GAs) and SA algorithms are found to have a better
global perspective than the traditional methods (Deb, 1996). Moreover,
when an optimization problem contains multiple global solutions, just
the best global optimum solution may not be the desirable one. It is
always prudent to know about other equally good solutions, which
correspond to a marginally inferior objective function values but more
amenable to be accepted. However, if the traditional methods are used to
find multiple optimal solutions, they need to be applied a number of
times, each time starting from a different initial guess and hoping to
achieve a different optimal solution. Simulated annealing (SA) is a
probabilistic optimization technique, which mimics the cooling
phenomenon of molten metals to constitute a search procedure
(Kirkpatrick et al., 1983). In an effort to predict the VLE of
C[O.sub.2] in aqueous amine solvents with better accuracy, Kundu et al.
(2003) and Kundu and Bandyopadhyay (2005, 2006a, 2006b) applied SA
successfully besides LM, in parameter estimation. In the present paper,
DE algorithms (Price and Storn, 1997) have been used for estimation of
interaction parameters of the VLE model over a wide range of
temperature, C[O.sub.2] partial pressure, and amine concentration range.
Differential evaluation (DE) is a generic name for a group of
algorithms, which is based on the principles of GA but have some
inherent advantages over GA, like its simple structure, ease of use,
speed, and robustness. A relative comparison among these three different
optimization algorithms has been made with respect to the VLE prediction
accuracy for the aforesaid systems. For (C[O.sub.2] + MDEA + [H.sub.2]O)
system DE performance is comparable to that of SA and LM performance.
For (C[O.sub.2] + AMP + [H.sub.2]O) system, DE seems to predict better
than SA and LM predictions. DE has been successfully applied in various
fields: dynamic optimization of a continuous polymer reactor, estimation
of heat transfer parameters in trickle bed reactor, optimal design of
heat exchangers, optimal design of shell and tube heat exchanger,
synthesis and optimization of heat integrated distillation systems,
expert systems for the optimal design of heat exchangers, etc., but to
the best of our knowledge this is the first ever application of DE in
phase equilibrium of acid gas + aqueous alkanolamine systems.
MODEL USED
Chemical Equilibria
In the aqueous phase of (C[O.sub.2] + MDEA + [H.sub.2]O) and
(C[O.sub.2] + AMP + [H.sub.2]O) systems the following chemical
equilibria are involved:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)
For MDEA, R, R', and R" are C[H.sub.3],
[C.sub.2][H.sub.4]OH, and [C.sub.2][H.sub.4]OH, respectively; for AMP R,
R', and R" are H, H, and [C.sub.2][H.sub.2](C[H.sub.3])2OH,
respectively. [K.sub.1], [K.sub.2], [K.sub.11], [K.sub.22], [K.sub.33]
are the chemical equilibrium constants taken from Austgen and Rochelle
(1991); Silkenbaumer et al. (1998); Li and Shen (1993) and are presented
in Table 1. All the equilibrium constants used in the present model are
mole fraction based after appropriate conversion, wherever necessary
(Kundu et al., 2003). Since the present work is limited to low to
moderate partial pressure range, the non-ideality of the vapour phase is
neglected. In the low to moderate range of C[O.sub.2] partial pressure
the fugacity of C[O.sub.2] is assumed to be its partial pressure and
solubility of C[O.sub.2] is identical to Henry's constant
(HC[O.sub.2]). The VLE of C[O.sub.2] over the aqueous alkanolamine
solvent, assuming no amine (solvent) species in the vapour phase, is
given as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)
where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is the
free acid gas mole fraction in the liquid phase. Free molecular
C[O.sub.2] concentration in the liquid phase is negligible in comparison
to the other species present in the liquid phase below the C[O.sub.2]
loading of 1.0. Henry's law constant is a strong function of
temperature and taken from Austgen and Rochelle (1991) and presented in
Table 1.
Thermodynamic Framework
In (C[O.sub.2] + MDEA +[H.sub.2]O) system, two neutral species,
MDEA and [H.sub.2]O, and two ionic species, [MDEAH.sup.+] and
H[CO.sup.-.sub.3] in the equilibrated liquid phase, have been
considered. There are four major species, two neutral solvents, AMP and
[H.sub.2]O, and two ionic species, [AMPH.sup.+] and HC[O.sup.-.sub.3] ,
in the equilibrated liquid phase for the (C[O.sub.2] + AMP + [H.sub.2]O)
system. 2-Amino-2-methyl-1- propanol (AMP), being a sterically hindered
primary amine, the AMP-carbamate formed is unstable and it may undergo
carbamate reversion reaction (Sartori and Savage, 1983). Yih and Shen
(1988) have presumed that sterically hindered amine; AMP cannot form
carbamate. Hence, the following alternative reaction was proposed by
them for AMP:
RR' N [H.sup.+] CO[O.sup.-] + [H.sub.2]O [left and right
arrow] RR' [NH.sup.+.sub.2] + HC[O.sup.-.sub.3] (7)
Another alternative mechanism for the bicarbonate formation has
been proposed by Chakraborty et al. (1986).
RR' NH + C[O.sub.2] + [H.sub.2]O [left and right arrow]
RR' N[H.sup.+.sub.2] + HC[O.sup.-.sub.3] (8)
However, Equation (8) is similar to reaction of C[O.sub.2] with
tertiary amines (e.g., MDEA). Hence, the existence of AMP carbamate in
the equilibriated liquid phase has been ruled out though it is a primary
alkanolamine. Since concentration of free molecular species C[O.sub.2]
and the ionic species C[O.sup.2-.sub.3], and [OH.sup.-] in the liquid
phase are very low compared to the other species present in the
equilibrated liquid phase, they have been neglected in the material
balances and activity coefficient calculations. For both the systems, it
has been assumed approximately that all the dissolved C[O.sub.2] in the
liquid phase is converted in to HC[O.sup.-.sub.3] ion. Several previous
workers, Li and Mather (1994, 1996, 1997); Haji-Sulaiman et al. (1996);
Posey (1996) have observed that neglecting their concentrations in the
liquid phase in this system for C[O.sub.2] loading below 1.0 does not
result in significant error in the VLE predictions. The standard state
associated with each solvent is the pure liquid at the system
temperature and pressure. The adopted standard state for ionic solutes
is the ideal, infinitely dilute aqueous solution (infinitely dilute in
solutes and alkanolamines) at the system temperature and pressure. The
reference state chosen for molecular solute C[O.sub.2] is the ideal,
infinitely dilute aqueous solution at the system temperature and
pressure. This leads to the following unsymmetric normalization of
activity coefficients:
Solvents : [[gamma].sub.s] [right arrow] 1 as [x.sub.s] [right
arrow] 1
Ionic and neutral solutes : [[gamma].sub.i] [right arrow] 1 as
[x.sub.i] [right arrow] 0, [x.sub.s[not equal to]w] = 0
where, s refers to any non-aqueous solvent, i refers to ionic or
neutral solute. Activity coefficients of all species are assumed to be
independent of pressure.
Thermodynamic Expression of Equilibrium Partial Pressure
The thermodynamic expression for equilibrium partial pressure of
C[O.sub.2] in aqueous MDEA solutions is as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (9)
where, [x.sub.n.sup.s] are the liquid phase mole fractions of the
components, based on true molecular or ionic species at equilibrium. The
calculation of the concentration for each component at equilibrium is as
follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (10)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (11)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (12)
Activity coefficients [[gamma].sub.i] (based on mole fraction
scale) for different species present in the liquid phase are calculated
from modified Clegg-Pitzer equation in the "activity coefficient
model". Similar expression for partial pressure of C[O.sub.2] in
aqueous AMP can also be written.
Activity Coefficient Model
The modified Clegg-Pitzer equations have been used to derive the
activity coefficients of different species present in the equilibrated
liquid phase (Li and Mather, 1994; Kundu et al., 2003; Kundu and
Bandyopadhyay, 2005). Differentiating the expressions for the
short-range and long-range force contribution to the excess Gibbs energy
one can get activity coefficients for the ions and the neutral species.
The expression for the activity coefficient for ionic solutes in the
aqueous amine solvent is as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (13)
where 1= [H.sub.2]O, 2 = AMP/MDEA, M = [AMPH.sup.+]/[MDEAH.sup.+],
X = HC[O.sup.-.sub.3].
For neutral molecules as for example, water:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (14)
where 1 = [H.sub.2]O, 2 = AMP/MDEA, M = [AMPH.sup.+]/[MDEAH.sup.+],
X = HC[O.sup.-sub.3]. Similar expressions for logarithm of activity
coefficient can be derived for other molecular and ionic solutes. The
function g(x)
where x = ([[alpha].sub.1][I.sup.1/2.sub.x]) = 2[I.sup.1/2] (16) in
Equation (13) is expressed by:
g(x) = 2 [1 - (1 + x) exp(-x)]/[x.sup.2] (15)
where x = ([[alpha].sub.1] [I.sup.1/2.sub.x]) = [2I.sup.1/2] (16)
DATA REGRESSION: ESTIMATION OF INTERACTION PARAMETER
In this work, the experimental solubility data of C[O.sub.2] in
aqueous MDEA in the loading range of 0.001-1.0 mol/mol, partial pressure
range of 0.01-5500 kPa, concentration ranging from 2.53-4.28 M,
temperature range of 298-393 K and experimental solubility data of
C[O.sub.2] in aqueous AMP in the loading range of 0.03-1.0 mol/mol,
concentration range of 2.8-3.43 M, temperature in the range 303-373 K,
and the partial pressure of 0.3-1000 kPa have been used to estimate the
interaction parameters by regression analysis. At first all available
experimental data from different authors were used for regression
analysis to obtain the interaction parameters, which resulted in a large
average correlation deviation. Then a lot of equilibrium curves were
made at the same temperatures and the same initial amine concentrations
but from the different authors and some sets of data, which were far
away from most of the data, were discarded. Finally, the combination of
data useful for generating a correlation to obtain a set of interaction
parameters has been identified. Since the adjustable interaction
parameters are characteristic of pair interactions of components of the
solution and are independent of solution composition, the fitted
parameters are valid outside the range of concentrations over which they
were fitted. Hence, the VLE model itself is valid at low acid gas
partial pressures even though the parameters of the activity coefficient
model were not fitted in this range (Kundu, 2004). The objective
function used for optimization is given by:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (17)
Among the interaction parameters for each system; one is for
ion-ion interactions, two are for ion-solvent interactions, and two are
for solvent-solvent interactions. The estimated interaction parameters
along with their temperature dependence and the temperature
coefficients, using DE technique for (C[O.sub.2] + [MDEA.sup.+]
[H.sub.2]O) and (C[O.sub.2] + AMP + [H.sub.2]O) systems are listed in
Tables 2 and 3, respectively. The objective function chosen in this work
takes care of rendering uniform weightage throughout the entire range of
partial pressure (from low to high), provided the data scatter
throughout the entire range of partial pressure (from low to high), as
claimed by the previous workers, is more or less uniform. In reality,
relatively small amounts of low-loading data are available compared to
the moderate/high-loading data, which are of rather poor precision, and
possibly of poor accuracy too (Weiland et al., 1993). The present
authors think that a higher weighing for the low partial pressure region
in the chosen objective function may not be entirely undesirable to
predict the low-pressure data with acceptable accuracy using the model,
in view of the immense significance of the low-pressure data for
gas-treating processes.
METHOD OF SOLUTION
Differential evolution (DE) technique is used to estimate the
interaction parameters for the VLE model. DE is a stochastic,
population-based method. These methods heuristically "mimic"
biological evolution, namely, the process of natural selection and the
"survival of the fittest" principle. An adaptive search
procedure based on a "population" of candidate solution points
is used. "NP" denotes the population size. In a population of
potential solutions within an n-dimensional search space, a fixed number
of vectors are randomly initialized, then evolved over time to explore
the search space and to locate the minima of the objective function.
"D" denotes the dimension of each vector, which is actually
the number of optimum parameters to be estimated of the proposed
objective function (Equation (17)).
The main operation in DE is the NP number of competitions, which
are to be carried out to decide the next generation population.
Generations or iterations involve a competitive selection that drops the
poorer solutions. From the current generation population of the vectors,
one target vector is selected. Among the remaining population vectors,
DE adds the weighted (weight factor is denoted by F, and is specified at
the starting) difference between two randomly chosen population vectors
to a third vector, called trial vector (randomly chosen), which results
in a "noisy" random vector. This operation is called
recombination (mutation). Subsequently, crossover is performed between
the trial vector and the noisy random vector (perturbed trial vector) to
decide upon the final trial vector or offspring of this generation. For
mutation and crossover to be carried out together, a random number is
generated which is less than the CR (crossover constant). If the random
number generated is greater than CR, then the vector taken for mutation
(trial vector) is kept copied as it is; as an offspring of this
generation (mutation is not compulsory). This way no separate
probability distribution has to be used which makes the scheme
completely self-organizing. Finally, the trial vector replaces the
target vector for the next generation population, if, and only if it
yields a reduced value of the objective function than in comparison to
the objective function based on target vector. In this way, all the NP
number of vectors of the current generation is selected one by one as
target vectors and checked whether the trial vector (offspring) to
create the population of the next generation should replace them or not.
The control parameters of the algorithm are: number of parents (NP),
weighing factor or mutation constant (F), crossover constant (CR). There
is always a convergence speed (lower F value) and robustness (higher NP
value) trade-off. CR is more like a fine tuning element. High values of
CR like CR=1 give faster convergence if convergence occurs. The flow
chart for the DE technique is given in Figure 1.
[FIGURE 1 OMITTED]
Different strategies can be adopted in DE algorithm depending upon
the type of problem for which DE is applied. The strategies can vary;
based on the vector to be perturbed, number of difference vectors
considered for perturbation, and finally the type of crossover used
(Babu, 2004). The DE/rand-best/1/bin strategy has been adopted here. The
general convention used above is DE/x/y/z. DE stands for differential
evolution, x represents a string denoting the vector to be perturbed, y
is the number of difference vectors considered for perturbation of x,
and z stands for the type of crossover being used (exp, exponential;
bin, binomial). The perturbation can be either in the best vector of the
previous generation or in any randomly chosen vector. Similarly for
perturbation either single or two vector differences can be used. In
exponential crossover, the crossover is performed on the D (the
dimension, i.e., number of variables to be optimized) variables in one
loop until it is within the CR bound. In binomial crossover, the
crossover is performed on each of the D variables whenever a randomly
picked number between 0 and 1 is within the CR value. DE/rand-best/1/bin
strategy has been adopted in the present problem.
Algorithm for Differential Evolution Technique
Step 1: Initialization of the parameters required for DE. Following
are the values considered in the present work, population size (NP)=50,
crossover (CR)=0.9, mutation constant (F)=0.8.
Step 2: Generation of the population randomly between the upper and
lower bounds of the desired parameters.
Step 3: Calculation of the objective function value for all
population vectors.
Step 4: Random selection of three population points a, b, c such
that these are not equal, followed by the generation of a random number,
if the generated random number is less than CR, leading to the mutation
according to Equation (18) followed by a crossover. X(c) is now mutated
to X(m).
X(m) = X(c) + F(X(a) - X(b)) (18)
with a check for bounds of the vector generated from the mutation
step. If bounds are violated, then generation of "X" randomly
between the bounds; otherwise X(m)=X.
Step 5: If the generated random number is greater than CR, mutation
is not necessary, hence, X(m)=X(c).
Step 6: Calculation of the objective function for X(m) vector.
Step 7: Selection of the least objective function value.
Step 8: Repetition of steps 4-6 for all the populations.
Step 9: Repetition of steps 4-7 until the termination criteria are
met.
Step 10: Stop.
The termination criteria taken in the present study are of two
types. First, if there is no improvement in objective functional value
for some generations, and second, the difference in the maximum
objective value and minimum objective value in the population is less
than [10.sup.-12]. The hardware platform used was an Intel Pentium IV processor, 1.4 GHz. The developed DE code using MATLAB 7.0 version had
to compromise between CPU time and the solution quality. For (C[O.sub.2]
+ [MDEA.sup.+] [H.sub.2]O) system, sometimes it took 10 000 generations
to reach a good quality solution.
RESULTS AND DISCUSSION
Three different kinds of optimization techniques; traditional and
gradient based (LM), probabilistic (SA), and evolutionary algorithm (DE)
have been used here for predicting phase equilibria. LM is a non-linear
least-squares method, which uses a steepest descent method and cannot
guarantee a global optimum solution; limited by the degree of
non-linearity and initial guess. SA is a generic probabilistic
meta-algorithm locating a good approximation to the global optima of a
given objective function in a search space. The main driving force
behind SA is to occasionally allow for wrong-way movement (uphill moves
for minimization) saving the method from becoming stuck at local optima.
DE maintains a pool of solutions rather than just one. New candidate
solutions are generated not only by "mutation" (as in SA), but
also by "combination" of two solutions from the pool. It
ensures to have a better global perspective of the solution than the
other two techniques discussed.
(C[O.sub.2] + [MDEA.sup.+] [H.sub.2] O) System
For (C[O.sub.2] + [MDEA.sup.+] [H.sub.2]O) system, two solvent-ion
pair interactions, one ion-ion interaction, and two solvent-solvent
Margules interaction parameters were determined by regression analysis
while using SA and DE technique. Li and Mather (1997) derived the
Margules interaction parameters by regressing binary (MDEA + [H.sub.2]O)
VLE data. Kundu and Bandyopadhyay (2005) used those solvent-solvent
interaction parameters without any further regression. Hence, two
solvent-ion pair interactions and one ion-ion interaction parameters
were determined by regression analysis while predicting VLE using LM
technique. The interaction parameters derived by using SA and LM
technique from our previous work (Kundu and Bandyopadhyay, 2005) have
been used here to compare the VLE prediction performances of three
different techniques.
The same data set belonging to the identical conditions of
temperature, pressure, and concentration ranges were considered for
regression analysis by different techniques, SA and LM, to determine the
interaction parameters of the VLE model. DE resulted in a higher
correlation and prediction deviations, while using those same data sets
for generation of the interaction parameters; hence different
combination of data sets are considered in DE technique (in regression
analysis) to generate the interaction parameters for (C[O.sub.2] +
[MDEA.sup.+ [H.sub.2]O) system. The temperature coefficients of the
interaction parameters have been fitted to the linear/polynomial
functionality of temperature with an average regression coefficient of
0.99 [+ or -] 0.03 after being estimated by the SA technique at closely
spaced interval of temperature. For the LM and DE techniques, the
regression analysis was done along with the temperature coefficients
relating the interaction parameters.
The experimental solubility data used for regression analysis by
DE, SA, and LM, to estimate the interaction parameters, along with their
correlation deviations have been summarized in Table 4. The deviation of
correlation varies in the range from 7.3 to 15.8% by SA, 10 to 40% by LM
whereas 7.32 to 14.52% by DE technique. In the DE technique, the average
absolute deviation between all the experimental and correlated
C[O.sub.2] partial pressure was 10.8%. The fitted interaction parameters
derived by DE (presented in Table 2), SA and LM technique (taken from
Kundu and Bandyopadhyay, 2005) for C[O.sub.2] - MDEA - [H.sub.2]O
ternary system have been used to predict some solubility data, which
were not used for regression (correlation). The prediction results using
DE technique are summarized in Table 5. Table 5 also presents a
description of VLE prediction accuracy of the same model using three
different optimization techniques, LM, SA, and DE, when compared with
the experimental results of seven different groups. Figure 2 is a
typical parity plot showing some of the predicted results. On a relative
assessment of using the LM and the SA algorithms for parameter
estimation, it was found earlier by Kundu et al. (2003) and Kundu and
Bandyopadhyay (2005) that use of the SA algorithm resulted in a
relatively better accuracy in correlation and prediction of VLE of
(C[O.sub.2] + MDEA + [H.sub.2]O) and (C[O.sub.2] - AMP- [H.sub.2]O)
system for most of the cases. In the present work, it has been found
that (Table 5), both SA and DE techniques have shown a comparable VLE
prediction accuracy for (C[O.sub.2] + MDEA + [H.sub.2]O) system, which
is better than LM for some data sets, considered earlier by Kundu and
Bandyopadhyay (2005). Figure 2 shows the typical parity plot showing
some of the predicted results for this system. It is to be mentioned
that the predicted C[O.sub.2] partial pressures using DE technique are
in excellent agreement with the data set of Kuranov et al. (1996) for
4.13 M MDEA solution over a temperature range of 313-413[degrees]C,
where the other two techniques did not perform appreciably well.
[FIGURE 2 OMITTED]
For (C[O.sub.2]-AMP-[H.sub.2]O) System
For (C[O.sub.2] + AMP +[H.sub.2]O) system, two solvent-ion pair
interactions, one ion-ion interaction, and two solvent-solvent Margules
interaction parameters are determined by regression analysis. The same
data set belonging to the identical conditions of temperature, pressure,
and concentration ranges were not considered for regression analysis by
different techniques, LM, SA, and DE in determining the interaction
parameters of the VLE model for this particular system. The interaction
parameters derived on the basis of same data sets for three optimization
techniques resulted in relatively higher correlation and prediction
deviation for any one or two of the techniques. Hence, different
combinations of data sets were considered for three different techniques
to generate the interaction parameters for (C[O.sub.2] + AMP
+[H.sub.2]O) system; which may lead to minimum possible correlation and
prediction deviations. The experimental solubility data used for
regression analysis by DE, SA, and LM, to estimate the interaction
parameters, along with their correlation deviations have been summarized
in Table 6. The deviation of correlation varies in the range from
5.67%-15.5% by DE, 9.34-16.0% SA, and 11.0-15.32% by the LM technique.
The average absolute deviation between all the experimental and
correlated C[O.sub.2] partial pressure was 8.77% by the DE technique.
The fitted interaction parameters derived by DE (presented in Table 3),
SA, and LM techniques (taken from Kundu et al., 2003) for (C[O.sub.2] +
AMP + [H.sub.2]O) ternary system have been used to predict some
solubility data, which were not used for regression (correlation). The
prediction results using DE technique are summarized in Table 7. Table 7
also presents a description of VLE prediction accuracy of the same model
using three different optimization techniques, LM, SA, DE, when compared
with the experimental results of five different groups. Figure 3 is a
typical parity plot showing some of the predicted results for this
system. It has been found that the prediction accuracy using DE
technique for (C[O.sub.2] + AMP + [H.sub.2]O) system is better than both
SA and LM when compared on the basis of the same data sets considered by
Kundu et al. (2003) for their relative assessment on VLE prediction
accuracy using SA and LM techniques. Figure 4 shows the relatively
better prediction ability of DE in comparison to the other two
techniques for (C[O.sub.2] + AMP + [H.sub.2]O) system, while compared
with the experimental results of Jane and Li (1997).
[FIGURE 3 OMITTED]
[FIGURE 4 OMITTED]
CONCLUSIONS
The modified Clegg-Pitzer equations have been used to correlate and
predict the VLE of (C[O.sub.2] + AMP + [H.sub.2]O) and (C[O.sub.2] +
MDEA + [H.sub.2]O) systems, which are of immense significance as far as
gas-treating processes are concerned. The present work requires solving
a multivariable optimization problem. DE algorithms have been used for
parameter estimation of the developed VLE model. In this work, the
DE/rand-best/1/bin strategy has been used. A relative comparison among
different traditional and non-traditional optimization algorithms has
been made with respect to the VLE prediction accuracy for the aforesaid
systems. For the (C[O.sub.2] + MDEA + [H.sub.2]O) system DE performance
is comparable to that of SA, and better than LM performance. For
(C[O.sub.2] + AMP + [H.sub.2]O) system, DE seems to predict
qualitatively better than SA and LM predictions.
ACKNOWLEDGEMENTS
The financial support by the Department of Science and Technology
(DST) and Centre for High Technology (CHT), New Delhi, India, are
gratefully acknowledged. The authors are grateful to Professor A. E.
Mather, University of Alberta, Canada, for providing some important
experimental VLE data for C[O.sub.2] in aqueous MDEA and AMP solutions.
NOMENCLATURE
[A.sub.x] Debye-Huckel parameter on a mole
fraction basis
[A.sub.nn'] interaction parameter between
neutral molecules
[B.sub.ca] ion-ion interaction parameter
[C.sup.0.sub.x] initial concentration of species
[F.sub.x] cationic fraction
[MATHEMATICAL EXPRESSION Henry's law constant for C[O.sub.2],
NOT REPRODUCIBLE IN ASCII] kPa
I ionic strength on a molar
concentration basis
[I.sub.x] ionic strength on mole fraction
basis
[K.sub.1], [K.sub.2], [K.sub.11], thermodynamic chemical equilibrium
[K.sub.22], [K.sub.33] constants
[MATHEMATICAL EXPRESSION partial pressure of C[O.sub.2], Pa
NOT REPRODUCIBLE IN ASCII] or kPa
T absolute temperature, K
W interaction parameter between
neutral and ionic species
Z valency of an ion
Greek Symbols
[[alpha].sub.1] Pitzer universal constant in Equation
(16)
[MATHEMATICAL EXPRESSION NOT liquid phase loading of C[O.sub.2], kmol
REPRODUCIBLE IN ASCII] C[O.sub.2/kmol amine
[gamma] activity coefficient
[psi] objective function for regression
Subscripts
a, X anion
c, M cation
n,n' neutral solvent species
cal calculated value
ex experimental value
s solvent
i ionic or neutral solute
w water solvent
Note in Equation (11) [alpha] is to be written as [MATHEMATICAL
EXPRESSION NOT REPRODUCIBLE IN ASCII] which is liquid phase loading of
C[O.sub.2], kmol C[O.sub.2]/kmol amine same as used in Equation (10).
Manuscript received October 20, 2006; revised manuscript received
April 4, 2007; accepted for publication April 10, 2007.
REFERENCES
Austgen, D. M. and G. T. Rochelle, "Model of Vapor-Liquid
Equilibria for Aqueous Acid Gas-Alkanolamine Systems. Representation of
[H.sub.2]S and C[O.sub.2] Solubility in Aqueous MDEA and C[O.sub.2]
Solubility in Aqueous Mixtures of MDEA with MEA or DEA," Ind. Eng.
Chem. Res. 30, 543-555 (1991).
Babu, B. V., "Process Plant Simulation," Oxford
University Press, New Delhi, India (2004).
Chakma, A. and A. Meisen, "Solubility of C[O.sub.2] in Aqueous
Methyldiethanolamine and N,N-Bis(hydroxyethyl) Piperazine Solutions," Ind. Eng. Chem. Res. 26, 2461-2466 (1987).
Chakraborty, A. K., G. Astarita and K. B. Bischoff,
"C[O.sub.2] Absorption in Aqueous Solutions of Hindered
Amines," Chem. Eng. Sci. 41, 997-1003 (1986).
Dawodu, O. F. and A. Meisen, "Solubility of Carbon Dioxide in
Aqueous Mixture of Alkanolamines," J. Chem. Eng. Data 39, 548-552
(1994).
Deb, K., "Optimization for Engineering Design: Algorithms and
Examples," Prentice-Hall, New Delhi (1996).
Deshmukh, R. D. and A. E. Mather, "A Mathematical Model for
Equilibrium Solubility of Hydrogen Sulfide and Carbon Dioxide in Aqueous
Alkanolamine Solutions," Chem. Eng. Sci. 36, 355-362 (1981).
Haji-Sulaiman, M. Z., M. K. Aroua and M. I. Pervez,
"Equilibrium Concentration Profiles of Species in
C[O.sub.2]-Alkanolamine-Water Systems," Gas Sep. Purif. 10, 13-18
(1996).
Jane, I. S. and M. H. Li, "Solubilities of Mixtures of Carbon
Dioxide and Hydrogen Sulfide in Water + Diethanolamine +
2-Amino-2-Methyl-1-Propanol," J. Chem. Eng. Data 42, 98-105 (1997).
Jou, F. Y., J. J. Caroll, A. E. Mather and F. D. Otto,
"Solubility of H2S and C[O.sub.2] in Aqueous Methyldiethanolamine
Solutions," Ind. Eng. Chem. Process Des. Dev. 21, 539-544 (1982).
Jou, F. Y., A. E. Mather and F. D. Otto, "Vapour- Liquid
Equilibrium of Carbon Dioxide in Aqueous Mixtures of Monoethanolamine
and Methyldiethanolamine," Ind. Eng. Chem. Res. 33, 2002-2005
(1994).
Jou, F. Y., J. J. Caroll, A. E. Mather and F. D. Otto, "The
Solubility of Carbon Dioxide and Hydrogen Sulfide in a 35 wt% Aqueous
Solution of Methyldiethanolamine," Can. J. Chem. Eng. 71, 264-268
(1995).
Kent, R. L. and B. Eisenberg, "Better Data for Amine
Treating," Hydrocarbon Process. 55, 87-90 (1976).
Kirkpatrick, S. C. D. Gelatt and M. P. Vechhi, "Optimization
by Simulated Annealing," Science 220, 671-680 (1983).
Kundu, M., "Vapour-Liquid Equilibria of C[O.sub.2] in Aqueous
Alkanolamines," Ph.D. Thesis, Indian Institute of Technology,
Kharagpur, India (2004).
Kundu, M. and S. S. Bandyopadhyay, "Modelling of Vapour-Liquid
Equilibrium of C[O.sub.2] in Aqueous N-Methyldiethanolamine Through the
Simulated Annealing Algorithm," Can. J. Chem. Eng. 83, 344-353
(2005).
Kundu, M. and S. S. Bandyopadhyay, "Vapour-Liquid Equilibrium
of C[O.sub.2] in Aqueous Blends of DEA/AMP," J. Chem. Eng. Data 51,
398-405 (2006a).
Kundu, M. and S. S. Bandyopadhyay, "Correlation and Prediction
of the Solubility of C[O.sub.2] in Water + Diethanolamine +
N-Methyldiethanolamine," Fluid Phase Equilib. 248, 158-167 (2006b).
Kundu, M., B. P. Mandal and S. S. Bandyopadhyay,
"Vapour-Liquid Equilibria of C[O.sub.2] in Aqueous Solutions of
2-Amino-2-Methyl-1-Propanol," J. Chem. Eng. Data 48, 789-796
(2003).
Kuranov, G., B. Rumpf, N. A. Smirnova and G. Maurer,
"Solubility of Single Gases Carbon Dioxide and Hydrogen Sulfide in
Aqueous Solutions of N-Methyldiethanolamine in the Temperature Range
313-413K at Pressures up to 5 MPA," Ind. Eng. Chem. Res. 35,
1959-1966 (1996).
Lemoine, B., Y. G. Li, R. Cadours, C. Bouallou and D. Richon,
"Partial Vapor Pressure of C[O.sub.2] and H2S over Aqueous
Methyldiethanolamine Solutions," Fluid Phase Equilib. 172, 261-277
(2000).
Li, M. H. and K. P. Shen, "Calculation of Equilibrium
Solubility of Carbon Dioxide in Aqueous Mixtures of Monoethanolamine
with Methyldiethanolamine," Fluid Phase Equilib. 85, 129-140
(1993).
Li, M. H. and B. C. Chang, "Solubilities of Carbon Dioxide in
Water + Monoethanolamine + 2-Amino-2-Methyl-1Propanol," J. Chem.
Eng. Data 39, 448-452 (1994).
Li, Y. G. and A. E. Mather, "Correlation and Prediction of the
Solubility of Carbon Dioxide in a Mixed Alkanolamine Solution,"
Ind. Eng. Chem. Res. 33, 2006-2015 (1994).
Li, Y. G. and A. E. Mather, "Correlation and Prediction of the
Solubility of C[O.sub.2] and H2S in Aqueous Solutions of
Triethanolamine," Ind. Eng. Chem. Res. 35, 4804-4809 (1996).
Li, Y. G. and A. E. Mather, "Correlation and Prediction of the
Solubility of C[O.sub.2] and H2S in Aqueous Solutions of
Methyldiethanolamine," Ind. Eng. Chem. Res. 36, 2760-2765 (1997).
Macgregor, R. J. and A. E. Mather, "Equilibrium Solubility of
H2S and C[O.sub.2] and Their Mixtures of Monoethanolamine and
Methyldiethanolamine," Can. J. Chem. Eng. 69, 1357-1366 (1991).
Posey, M. L. "Thermodynamic Model for Acid Gas Loaded Aqueous
Alkanolamine Solutions," Ph.D. Thesis, University of Texas, Austin,
TX, U.S.A. (1996).
Price, K. and R. Storn, "Differential Evolution," Dr.
Dobb's J., 18-24 (1997).
Roberts, B. E., "Solubility of C[O.sub.2] and H2S in Chemical
and Mixed Solvents," M.Sc. Thesis, University of Alberta, Edmonton,
AB, Canada (1983).
Sartori, G. and D. W. Savage, "Sterically Hindered Amines for
C[O.sub.2] Removal from Gases," Ind. Eng. Chem. Fundam. 22, 239-249
(1983).
Seo, D. J. and W. H. Hong, "Solubilities of Carbon Dioxide in
Aqueous Mixtures of Diethanolamine and
2-Amino-2-Methyl-1-Propanol," J. Chem. Eng. Data 41, 258-260
(1996).
Silkenbaumer, D., B. Rumpf and R. N. Lichtenthaler,
"Solubility of Carbon Dioxide in Aqueous Solutions of
2-Amino-2-Methyl-1-Propanol and N- Methyldiethanolamine and Their
Mixtures in the Temperature Range from 313 to 353 K and Pressure up to
2.7 MPA," Ind. Eng. Chem. Res. 37, 3133-3141 (1998).
Teng, T. T. and A. E. Mather, "Solubility of H2S, C[O.sub.2]
and Their Mixtures in an AMP Solution," Can. J. Chem. Eng. 67,
846-850 (1989).
Weiland, R. H., T. Chakravarty and A. E. Mather, "Solubility
of Carbon Dioxide and Hydrogen Sulfide in Aqueous Alkanolamines,"
Ind. Eng. Chem. Res. 32, 1419-1430 (1993).
Yih, S. M. and K. P. Shen, "Kinetics of Carbon Dioxide
Reaction with Sterically Hindered 2-Amino-2-Methyl-1-Propanol Aqueous
Solutions," Ind. Eng. Chem. Res. 27, 2237-2241 (1988).
M. Kundu (1) *, A. Chitturi (1) and S. S. Bandyopadhyay (2)
(1.) Chemical Engineering Group, Birla Institute of Technology and
Science, Pilani, Rajasthan, India
(2.) Separation Science Laboratory, Cryogenic Engineering Centre,
Indian Institute of Technology, Kharagpur, India
* Author to whom correspondence may be addressed. E-mail address:
mkundu@bits-pilani.ac.in
DOI 10.1002/cjce.20008
Table 1. Temperature dependence of the equilibrium constants and
Henry's constant
[K.sub.i] = exp [A + B / (T/K) + C ln(T/K) + D(T/K)], where i = 1,
2 and ii = 11 and 22, [K.sub.ii] = exp [A / (T/K) + B ln(T/K)
+ C(T/K) + D],
[K.sub.33] = exp [ A + B / (T/K) + C / [(T/K).sup.2] + D /
[(T/K).sup.3] + E / [(T/K).sup.4], H = exp [A (B / (T/K)) + C ln(T/K) +
D(T/K)]
Reaction Compound A B I
1 C[O.sub.2] 231.465 -12092.1 -36.7816
2 MDEA -9.4165 -4234.98 0
3 C[O.sub.2] -7742.6 -14.506 -0.028104
4 AMP -7261.78 -22.4773 0
5 [H.sub.2]O 39.5554 -9.879e4 0.568827e8
Henry's C[O.sub.2] 170.7126 -8477.711 -21.95743
constant
Reaction D E Reference
1 0 Austgen and Rochelle (1991)
2 0 Austgen and Rochelle (1991)
3 102.28 Silkenbaumer et al. (1998)
4 142.58612 Silkenbaumer et al. (1998)
5 -0.14645e11 0.136145e13 Li and Shen (1993)
Henry's 0.005781 Austgen and Rochelle (1991)
constant
Table 2. Estimated interaction parameters derived by the DE technique
for (C[O.sub.2] + MDEA + [H.sub.2]O) system B(or W or A) = a + b(T/K)
B or W or A A B
[B.sub.MX] 745.8125881999431 -2.02572012847313
[W.sub.1MX] 6.16389684502044 -0.00243179324964
[W.sub.2MX] -0.46731860042498 -0.05142212181018
[A.sub.12] 9.48748854648508 -0.02933601022959
[A.sub.21] 9.46640372458580 -0.02926811991783
1 = [H.sub.2]O, 2 = MDEA, M = [MDEAH.sup.+], X = HC[O.sub.3.sup.-]
Table 3. Estimated interaction parameters derived by the DE technique
for (C[O.sub.2] + MDEA + [H.sub.2]O) system B(or W or A) = a + b(T/K)
B or W or A A B
[B.sub.MX] -726.9140414231877 2.04601638504848
[W.sub.1MX] 155.9936324496222 -0.60069767840103
[W.sub.2MX] 240.7131667119178 -0.87405495529079
[A.sub.12] 471.7604672732886 -1.72023014237471
[A.sub.21] -46.58956159671227 0.13907473167950
1 = AMP, 2 = [H.sub.2]O, M = [AMPH.sup.+], X = HC[O.sub.3.sup.-]
Table 4. Experimental solubility data used in the regression analysis
to estimate the interaction parameters and the correlation deviations
for (C[O.sub.2]-MDEA-[H.sub.2]O) system
Reference [MDEA] Temperature (K)
(wt%)
Jou et al. (1994) 30 298, 313, 353, 393
Austgen and Rochell e (1991) 23.8 313
Jou et al. (1995) 35.0 313, 373
Chakma and Meisen (1987) 19.8 453
473
Chakma and Meisen (1987) 48.8 413
Dawodu and Meisen (1994) 48.8 373
Jou et al. (1994) 30 298
Silkenbaumer et al. (1998) 31.0 313
Jou et al. (1995) 35.0 373
Jou et al. (1982) 48.8 393
Reference C[O.sub.2] partial Data
pressure (kPa) points
Jou et al. (1994) 0.08-5775 54
Austgen and Rochell e (1991) 6.73-98.8 11
Jou et al. (1995) 0.719-263 18
Chakma and Meisen (1987) 138-4137 8
414-3585 6
Chakma and Meisen (1987) 138-4516 7
Dawodu and Meisen (1994) 558-3611 9
Jou et al. (1994) 0.1-176.95 13
Silkenbaumer et al. (1998) 12.0-86.8 10
Jou et al. (1995) 95.58-191 8
Jou et al. (1982) 0.143-5290 9
AAD% correlation (a)
Reference
SA DE LM
Jou et al. (1994) 11.9 -- 12.9
Austgen and Rochell e (1991) 12 -- 17.0
Jou et al. (1995) 15.8 -- 18.9
Chakma and Meisen (1987) 10.9 -- 15.7
9.3 40.7
Chakma and Meisen (1987) 15.7 -- 17.4
Dawodu and Meisen (1994) 7.3 -- 10
Jou et al. (1994) -- 14.3 --
Silkenbaumer et al. (1998) -- 9.38 --
Jou et al. (1995) -- 14.5 --
Jou et al. (1982) -- 7.32 --
(a) AAD% = [[summation over (n)] ([p.sub.cal] - [p.sub.exp]) /
[p.sub.exp]] / n x 100
Table 5. VLE prediction and comparison among different techniques for
(C[O.sub.2] + MDEA + [H.sub.2]O) system
Reference [MDEA] Temperature (K)
wt%
Jou et al. (1994) 30.0 313, 353, 393
Austgen and Rochelle (1991) 23.8 313
Jou et al. (1995) 35.0 313
Dawodu and Meisen (1994) 48.8 373
Macgregor and Mather (1991) 23.8 313
Jou et al. (1982) 23.8 313
Jou et al. (1982) 48.8 313, 343, 373
Chakma and Meisen (1987) 48.8 373
Kundu (2004) 23.8 303
30.0 313
323
Lemoine et al. (2000) 23.6 298
Kuranov et al. (1996) 47.0 313-413
Reference Data C[O.sub.2] partial
points pressure range (kPa)
Jou et al. (1994) 41 0.494-5775
Austgen and Rochelle (1991) 11 6.73-98.8
Jou et al. (1995) 10 0.719-100
Dawodu and Meisen (1994) 9 558-3611
Macgregor and Mather (1991) 4 1.17-588
Jou et al. (1982) 3 2.38-101
Jou et al. (1982) 34 200-4000
Chakma and Meisen (1987) 12 463-3600
Kundu (2004) 24 2-100
19 2-100
Lemoine et al. (2000) 13 0.02-1.636
Kuranov et al. (1996) 29 79-4460
Reference AAD% prediction (a)
DE SA LM
Jou et al. (1994) 26 --
Austgen and Rochelle (1991) 16.15 -- --
Jou et al. (1995) 36.6 -- --
Dawodu and Meisen (1994) 29.73 -- --
Macgregor and Mather (1991) 10.42 15.1 17.2
Jou et al. (1982) 30.66 32.4 27.4
Jou et al. (1982) 26.63 31.0 29.0
Chakma and Meisen (1987) 26.28 10.7 12.6
Kundu (2004) 19.69 13.6 23.7
20.75 10.1 19.0
Lemoine et al. (2000) 36.63 11.0 25.4
Kuranov et al. (1996) 13.31 34.0 36.0
(a) AAD% = [summation over (n)] ([p.sub.cal] - [p.sub.exp]) /
[p.sub.exp] / n x 100
Table 6. Experimental solubility data used in the regression analysis
to estimate the interaction parameters and the correlation deviations
for (C[O.sub.2]-AMP-[H.sub.2]O) system
Reference [AMP] Temperature (K) C[O.sub.2] partial
wt% pressure (kPa)
Teng and Mather (1989) 18.0 313,343 0.3-500
Li and Chang (1994) 30.0 313,333,353,373 1-200
Roberts (1983) 18.0 373 8.5-866
Kundu et al. (2003) 25.0 313 3.25-91.5
Seo and Hong (1996) 30.0 333 5.9-336
Li and Chang (1994) 30.0 373 1.12-71.2
Teng and Mather (1989) 18.0 343 2.43-434.6
Reference Data AAD% correlation (a)
points
SA DE LM
Teng and Mather (1989) 24 16.0 -- 11.0
Li and Chang (1994) 36 9.34 -- 12.55
Roberts (1983) 9 10.08 -- 15.8
Kundu et al. (2003) 8 -- 7.28 --
Seo and Hong (1996) 6 -- 5.67 --
Li and Chang (1994) 9 -- 15.5 --
Teng and Mather (1989) 9 -- 10.03 --
(a) AAD% = [[summation over (n)] ([p.sub.cal] - [p.sub.exp]) /
[p.sub.exp]] / n x 100
Table 7. VLE prediction and comparison among different techniques for
(C[O.sub.2] + AMP + [H.sub.2]O) system
Reference [AMP] Temperature
wt% (K)
Jane and Li (1997) 18.0 313
Roberts (1983) 18.0 313
18.0 373
28.5 313
Seo and Hong (1996) 30.0 313
353
Kundu et al. (2003) 18.0 303
25.0 303
323
30.0 303
313
323
Li and Chang (1994) 30.0 313
333
353
Reference Data C[O.sub.2] partial
points pressure range (kPa)
Jane and Li (1997) 7 1.45-84.2
Roberts (1983) 5 2.17-95.4
9 8.5-886
14 1.25-359
Seo and Hong (1996) 13 3.94-336
Kundu et al. (2003) 48 3.2-94
Li and Chang (1994) 27 1.05-197
Reference AAD% prediction (a)
DE SA LM
Jane and Li (1997) 12.99 20.0 23.5
Roberts (1983) 10.86 24.2 15.75
17.6 -- --
20.89 16.04 17.04
Seo and Hong (1996) 16.19 17.7 19.2
Kundu et al. (2003) 16.53 18.3 19.0
Li and Chang (1994) 23.37 -- --
(a) AAD% = [[summation over (n)] ([p.sub.cal] - [p.sub.exp]) /
[p.sub.exp]] / n x 100