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

net/publication/43996212

Bubble and dew point calculations in multicomponent and multireactive

mixtures

Article  in  Chemical and Biochemical Engineering Quarterly · September 2006

Source: OAI

CITATIONS READS

8 8,866

4 authors, including:

Adrian Bonilla-Petriciolet Juan Gabriel Segovia-Hernandez

Tecnologico Nacional de Mexico Universidad de Guanajuato
334 PUBLICATIONS   5,051 CITATIONS    262 PUBLICATIONS   2,629 CITATIONS   

SEE PROFILE SEE PROFILE

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

Statistical physics modeling of phosphate adsorption onto chemically modified carbonaceous clay View project

Book: Adsorption Processes for Water Treatment and Purification View project

All content following this page was uploaded by Adrian Bonilla-Petriciolet on 04 June 2014.

The user has requested enhancement of the downloaded file.

A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006) 111

BubbleandDewPointCalculationsinMulticomponentandMultireactiveMixtures
A. Bonilla-Petriciolet,* A. Acosta-Martínez, U. I. Bravo-Sánchez, and J. G. Segovia-Hernández**
Instituto Tecnológico de Aguascalientes, Departamento de Ingeniería Química,
Av. López Mateos 1801, CP 20256, Aguascalientes, Ags. México Original scientific paper
**Universidad de Guanajuato, Facultad de Ciencias Químicas, Received: August 28, 2005
Noria Alta s/n, C.P. 36050, Guanajuato, Gto. México Accepted: February 1, 2006

Bubble and dew point calculations are useful in chemical engineering and play an
important role in the study of separation equipments for non-reactive and reactive mix-
tures. To the best of the authors’s knowledge, few methods have been proposed for these
calculations in systems with several chemical reactions. The objective of this paper is to
introduce new conditions for performing bubble and dew point calculations in reactive
mixtures. We have developed these conditions based on the application of transformed
variables of Ung and Doherty (1995). Using these transformed variables, the solution
space is restricted to compositions that are already at chemical equilibrium and by conse-
quence the problem dimension is also reduced. The reliability and efficiency of three
equation-solving methods are tested and compared using our equilibrium conditions: a) a
simultaneous equation-solving approach using Newton method (SESN), b) an equa-
tion-decoupling approach using successive substitution method (EDSS) and c) an opti-
mization approach using the stochastic optimization method Simulated Annealing
(OSA). Our results indicated that even for simple reactive systems, bubble and dew point
calculations are challenging for classical equation-solving methods and require robust
strategies. We conclude that OSA and EDSS methods are reliable to locate bubble and
dew points in reactive systems. EDSS is more efficient than OSA; however, OSA does
not need initial guesses and is more suitable for difficult problems.
Key words:
Chemical equilibrium, phase equilibrium, bubble point, dew point, global optimization,
equation-solving method comparison

Introduction A special case of phase equilibrium problems

are the bubble and dew point calculations. These
The description of phase equilibrium in calculations are useful in chemical engineering and
multireaction mixtures is an important topic in- play an important role in the study of separation
volved in several industrial applications and it is the equipments.14 Under this context, robust, and effi-
basis for design, analysis, and synthesis of separa- cient methods for these calculations are desirable.
tion processes.1 The phase equilibrium problem To the best of the authors’s knowledge, few meth-
with several chemical reactions is highly non linear ods have been developed to calculate bubble and
and multivariable.2 During the last years, there has dew points in multireactive mixtures.9,12
been a growing interest for developing new meth-
This paper introduces a new method for per-
ods to treatment the thermodynamic behavior
forming dew and bubble point calculations in
of mixtures under physical and chemical equilib-
multireactive and multicomponent systems. We have
rium.2–13 Almost all developed methods use the
proposed new conditions for these equilibrium prob-
mole fractions and reaction extents as independent
lems based on the theory of reactive variables of
variables and unknowns of the reactive phase equi-
Ung and Doherty.13 Also, we have tested and com-
librium problem. Until now, only a few methods
pared three equation-solving methods for performing
have used theories of variable transformation with
those calculations, using our equilibrium conditions.
the aim of reducing problem dimensionality and
improving the numerical behavior (efficiency and
reliability) of solution methods.1,9 These kinds of
algorithms are very attractive for the simulation of Formulation of new conditions
separation process and favor the study of complex for bubble and dew point calculations
multireactive multicomponent systems. in multireactive mixtures

*Corresponding author: Ung and Doherty13 showed that the Gibbs en-
(52) 4499105002 ext. 127, [email protected] ergy in a reactive system behaves as in a non-reac-
112 A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006)

tive system if transformed composition variables and the transformed mole fractions x$ i are given
are used instead of the conventional composition by8,9,13
variables. Using transformed compositions, we re-
strict the solution space to the compositions that are x i - v i N -1 x ref
x$ i = 0 i = 1,¼ , n - n R (6)
already in chemical equilibrium. This reduces the 1- v TOT N -1 x ref
problem dimension by the number of independent
reactions and makes it possible to represent phase where xref is the column vector of nR reference com-
equilibrium in reactive systems in a similar way as ponent mole fractions, vi is the row vector of
in non-reactive mixtures. So, the reactive phase dia- stoichiometric number of component i for each re-
grams look similar to the non-reactive ones and the action vTOT, is a row vector where each element cor-
non-reactive flash algorithms can be easily modi- responds to reaction r and it is the sum of the
fied to account for the equilibrium reactions.12 stoichiometric number for all components that par-
Also, Ung and Doherty13 showed that the chemical ticipate in reaction r, and N is a square matrix
potential follows all the thermodynamic relation- formed from the stoichiometric number of the refer-
ships of a non-reactive system as long as all the ence components in the nR reactions.8,9,13 The refer-
thermodynamic properties are functions of the ence mole fractions are calculated using Eq (6) and
transformed composition variables. Considering from the equilibrium constants for each reaction
this, equilibrium conditions among two phases of a K ¢eq by solving a system of nR nonlinear equations
reactive mixture, with n components and nR inde- given by
pendent chemical reactions, are met when
n
vr
$ ai
Dm $ ib
Dm K ¢eq = Õ a i i r = 1,¼ , n R (7)
= i = 1,¼ , n - n R (1) i=1
RT RT
where ai is the activity of component i and v ir is the
where
stoichiometric number of component i in reaction r,
$i
Dm
j
$ i {x$ }- m i0
Dm
j respectively. When we know the reference mole
= (2) fractions, the remaining mole fractions are calcu-
RT RT lated using Eq (6).
We write the phase equilibrium condition in
with R the universal gas constant, m 0i the chemical terms of transformed variables, using Eq (4) – (6),
potential of the pure component and m $ ij the trans- as follows
formed chemical potential of component i at phase j b
which is a function of the transformed mole fraction x$ i = x$ ia K$ i q + d i i = 1,¼ , n - n R (8)
x$ . Eq (2) can be expressed in terms of fugacity or
activity coefficients as follows and

D$m i æ xi j
$ i {x$ }ö 1- v TOT N -1 x ref
a

= lnç
ç
÷= ln ( x y {x$ }) i=1, ¼ , n- n (3)
÷ q= b
(9)
RT è ji ø
i i R
1- v TOT N -1 x ref
b
where xi is the mole fraction of component i, j$ i is v i N -1 ( K$ i x ref
a
- x ref )
the fugacity coefficient of component i in the mix- di = b
i = 1,¼ , n - n R (10)
1- v TOT N -1 x ref
ture, ji is the fugacity coefficient of the pure com-
ponent, and gi is the activity coefficient of compo-
The material balance on component i is
nent i, respectively. Using Eq (1) and (3), we can
deduce the following relation for mole fractions at b b b
x$ F .i - (1- x$ F ) x$ ia - x$ F x$ i = 0 i=1,¼ , n - n R (11)
phase equilibrium
b
æ j$ a ö æg a ö and the transformed mole phase fraction x$ F is de-
b ç i ÷ ç i ÷
x i = x iaç b ÷= x iaç b ÷ i = 1,¼ , n - n R (4) fined as
è j$ i ø èg i ø
j j
j x F (1- v TOT N -1 x ref )
Then, we define the transformed phase equilib- x$ F = j = a, b (12)
(1- v TOT N -1 x Fref )
rium constant K$ i as
j
where x F is the conventional mole fraction of
j$ ai g ai
K$ i {x$ a , x$ b }= b
= b
i = 1,¼ , n - n R (5) phase j whose feasible domain is (0, 1) and xFref is
j$ i gi the column vector of nR reference component mole
A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006) 113

fractions in the initial overall composition. Using and dew points are calculated by solving simulta-
Eq (8) – (11), x$ ia is given by neously the following n – nR equations
b
b
x$ F , i - x$ F Fi = x$ i - x$ ia K$ i q - d i i = 1,¼ , n - n R - 1 (18)
x$ ia = b
i = 1,¼ , n - n R (13)
1+ x$ F ( K$ i q - 1) Fn-nR = f F b (19)

From the transformed mole fraction restriction where f F b is the bubble or dew point function,
n-nR Eq (16) or (17), respectively. The unknowns are
å x$ i, j = 1, we have n – nR – 1 transformed mole fractions of vapor or
i=1 liquid phase and the temperature or pressure. In this
work, the above equations are solved using the Nu-
n-nRé b ù
K$ i q ( x$ F , i - d i x$ F ) merical Recipes Fortran subroutine NEWT.
åê ê
b $ b
+ d i ú- 1= 0
ú
(14)
i=1 ë 1- $
x F + K i q x$ F û Equation-decoupling with Successive
Substitution method (EDSS)
n-nRé
x$ F , i - x$ F d i ù
b
This approach is often described in textbooks
åê b $
ê ú- 1= 0
ú
(15) on thermodynamics for flash calculations in non-re-
i=1 ë 1+ $
x F ( K i q - 1) û
active mixtures and is based on the Rachford-Rice
b b formulation. This method is reliable for flash calcu-
By applying the conditions x$ F = 0 and x$ F =1, lations in non-reactive systems and has a linear rate
we obtain the following new functions for bubble of convergence.16 The procedure used for bubble
and dew point calculations in multireactive mix- and dew point calculations in reactive systems is as
tures follows: in the inner loop, we have applied a New-
n-nR ton method, using the Numerical Recipes Fortran
subroutine NEWT, to update the temperature or
f bubble = 1- å ( K$ i q x$ F, i + d i ) = 0 (16)
pressure by solving Eq (16) or (17), while in the
i=1
outer loop the transformed mole fractions are up-
n-nR
é x$ F - d i ù dated with a successive substitution method using
f dew = 1- å êë ú= 0
K$ i q û
(17) Eq (8) or (13). In this work, we have not applied an
i=1 acceleration technique for this approach.
In contrast with other formulations,11,15 our Optimization approach using Simulated
functions are independent of reaction extents and Annealing method (OSA)
have fewer unknowns. Eq (16) and (17) are a func-
tion of temperature or pressure and n – nR trans- We have tested an optimization technique for
formed mole fraction x$ i of the liquid or vapor solving our equilibrium conditions. As indicated by
phase, respectively. On the other hand, these equi- Henderson et al.,17 the formulation of thermody-
librium problems are nonlinear and multivariable so namic calculations for optimization problems offers
that conventional numerical methods may present some advantages: a) the use of a robust optimiza-
difficulties, such as poor initialization or divergence tion method, b) the possibility of using a direct op-
behavior. We have performed a comparison of three timization method which requires only calculations
approaches to solve our bubble and dew point func- of the objective function and c) the use of an itera-
tions. In the next section we describe the strategies tive procedure whose convergence is independent
applied in this work. on the initial guess. Thus, the calculation of bubble
and dew point conditions can be performed by min-
imizing the next objective function
Solution approachs n-nR

Simultaneous equation-solving f obj = f F2 b + å Fi2 (20)

with Newton method (SESN) i=1

This is the classical approach used for perform- where Fi is described by Eq (18) for i = 1,…, n – nR
ing flash calculations in non-reactive and reactive and f F b is the bubble or dew point function, re-
systems since it is conceptually simple and straight- spectively. At the bubble and dew point conditions,
forward. The popular Newton method is used be- the global minimum of the objective function must
cause it provides quadratic convergence when the be zero, but we assume that a solution is found
initial estimates are close to the solution and it is when we find the transformed mole fractions and
readily available in computer programs.16 Bubble temperature or pressure that make the value of the
114 A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006)

objective function less than 1 · 10–6. The stochastic considered a tolerance of 1 · 10–6 for the conver-
global optimization method Simulated Annealing gence of all methods. All examples are solved 25
(SA) is used to minimize our objective function. SA times to evaluate the reliability and efficiency of
is inspired in the thermodynamic process of cooling solving methods, using in each time different ran-
of molten metals to achieve the lowest free energy dom initial values for unknowns x$ i , T or p. All cal-
state.18 The SA algorithm presents ease of computa- culations were performed on a Processor Intel
tional implementation and, if the values for its algo- Centrino 1600 MHz with 1.00 GB of RAM. The
rithm parameters are properly selected, it can con- performance of the three equation-solving methods
verge to the global optimum independently of ini- is tested using the following criterions: a) success
tial guesses. This optimization method has been rate of finding the bubble or dew point in the per-
successfully used in the resolution of several ther- $
formed calculations, b) average number of K-value
modynamic problems.17,19–24 It is important to note evaluations and c) computation time.
that other phase equilibrium problems (e.g. calcula- Phase stability of all calculated bubble and dew
tions of critical points and homogeneous azeo- points is evaluated by minimizing the tangent plane
tropes) have been solved using an optimization ap- distance function for reactive mixtures12
proach coupled with the SA method.17,24
n-nR
In our calculations, we have used the algorithm
proposed by Corana et al.25 because of its high reli-
RTPDF = å x$ i ( m$ i {x$}- m$ i {x$ F }) (21)
i=1
ability in thermodynamic calculations. In this algo-
rithm, a trial point is randomly chosen within the The RTPDF function is globally minimized us-
step length VM (a vector of length n variables) of a ing the Simulated Annealing method. Bonilla-Petri-
starting point defined in the feasible domain of op- ciolet24 has tested the SA method with several reac-
timization variables. The function is evaluated at tive systems and found that this method is robust
this trial point and its value is compared to its value for reactive phase stability analysis. All calculated
at the initial point. The Metropolis criterion,26 with bubble and dew points are stable.
a parameter called annealing temperature T, is used
to accept or reject the trial point. If the trial point is Example 1.
accepted, the algorithm moves on from that point. If isobutene + methanol « methyl ter-butyl ether
it is rejected, another point is chosen instead for a
trial evaluation. Each element of VM is periodically Our first example is the equilibrium reaction of
adjusted so that half of all function evaluations in that isobutene (1) and methanol (2) to produce methyl
direction are accepted. A fall in T is imposed upon ter-butyl ether (3), A1 + A2 « A3. This system has
the system with the R T variable by Tj+1 = R T × Tj, been analyzed by Okasinski and Doherty,28 Maier et
where j is the iteration counter and R T is the tem- al.29 and Harding and Floudas30 in the context of
perature reduction factor. Thus, as T declines, calculation of reactive homogeneous azeotropes.
downhill moves are less likely to be accepted and We have considered non-ideality in the liquid phase
the percentage of rejections rises. Given the scheme and ideal gas in the vapor phase. We use the Wilson
for the selection for VM, VM falls. Thus, as T de- equation for the activity coefficient in the liquid
clines, VM falls and SA focuses upon the most phase and the Antoine equation to calculate the
promising area for optimization. A full description properties of vapor phase. Maier et al.29 report the
of this algorithm is found in Corana et al.,25 and the parameters for both equations. The reaction equilib-
FORTRAN subroutine implemented by Goffe et rium constant is assumed to be independent of tem-
al.27 is used in this work. perature Keq = 0.04, and methyl ter-butyl ether
The choice of the cooling schedule is a crucial (MTBE) is the reference component so that the
aspect in the implementation of SA because it affects x1 + x 3
transformed mole fractions are x$ 1 = and
the numerical performance of the optimization pro- 1+ x 3
cedure. Based on preliminary calculations, we pro- x2 + x3
x$ 2 = = 1- x$ 1 .. We use x$ 1 and temperature
pose the following values for the quantities of SA: T0 1+ x 3
= 10.0, R T = 0.85 and NT = 5, where NT is the itera- as unknowns for all equation-solving methods. The
tion number before temperature reduction. initial values are randomly generated in the feasible
domains of (0, 1) for x$ 1 and (0.01, 300) °C for tem-
perature. These domains are also used in the imple-
Results and discussion mentation of the OSA method. Bubble and dew
point calculations are performed at different pres-
For the present study, four examples are used sures for a feed of x$ F (0.5, 0.5) and the results are
to compare the numerical performance of the three reported in Tab 1. The numerical performance of
equation-solving methods. In all examples, we have tested methods appears in Tab 2. In all calculations,
A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006) 115

T a b l e 1 – Bubble and dew point calculations for isobutene + methanol « methyl ter-butyl ether. Wilson equation and ideal gas,
Keq = 0.04 and x$F (0.5, 0.5).

Dew point Bubble point

p / atm
T / °C x$a T / °C x$ b

2 64.943 (0.0196, 0.9804) 13.944 (0.96245, 0.03755)

4 83.894 (0.02741, 0.97259) 36.425 (0.94082, 0.05918)

6 96.156 (0.03353, 0.96647) 51.356 (0.92324, 0.07676)

8 105.462 (0.03882, 0.96118) 62.859 (0.90804, 0.09196)

T a b l e 2 – Performance of equation-solving methods tested in bubble and dew point calculations in reactive systems
$
Success rate in percent (No. of K-values evaluations)1

Bubble point Dew point

Example No. T or p
SESN DESS OSA SESN DESS OSA

2atm 16 (73) 100 (56) 100 (21089) 16 (30) 100 (142) 100 (21345)

4atm 44 (72) 100 (52) 100 (21409) 36 (54) 100 (59) 100 (21329)
1
6atm 44 (84) 100 (48) 100 (21265) 20 (33) 100 (102) 100 (21145)

8atm 60 (84) 100 (43) 100 (21289) 24 (44) 100 (106) 100 (21273)

50°C 68 (55) 100 (371) 96 (42977) 4 (22) 100 (379) 100 (43297)

2 70°C 76 (53) 100 (247) 100 (43345) 20 (32) 100 (275) 100 (43377)

90°C 68 (59) 100 (208) 100 (43297) 40 (56) 100 (220) 100 (43505)

1atm 76 (86) 100 (270) 100 (32341) 4 (22) 100 (591) 100 (32677)

3 5atm 28 (50) 100 (329) 100 (31861) 12 (29) 100 (420) 100 (32233)

10atm 64 (64) 100 (370) 100 (32005) 8 (22) 100 (300) 100 (32449)

1atm 28 (20) 88 (92) 100 (21337) 28 (23) 100 (49) 100 (21217)

4atm 36 (16) 80 (46) 100 (21257) 28 (24) 100 (33) 100 (21241)
4
6atm 28 (15) 56 (40) 100 (21305) 56 (23) 100 (27) 100 (21417)

8atm 28 (13) 48 (35) 100 (21169) 72 (24) 100 (35) 100 (21433)
1Both parameters are calculated based on 25 calculations with random initial values.

DESS and OSA methods find the bubble and dew Example 2.
points without problems while SESN diverged sev- isobutene + methanol « methyl ter-butyl ether with
eral times. We observe that SESN shows more fail- n-butane as inert
ures in dew point calculations. These results indi-
cated, that even for simple reactive systems, bubble In the second example, we use the same reac-
and dew point calculations are challenging for clas- tion as before but include n-butane (4) as an inert.
sical equation-solving methods and require robust Ung and Doherty,9 Harding and Floudas30 and
$
strategies. The number of K-value evaluations of Maier et al.29 have also studied this reactive equi-
solving-methods is given by: OSA >>> DESS > librium problem. The transformed mole fractions
SESN. Computation time maintains around of 6 s x1 + x 3 x2 + x3
for this mixture are x$ 1 = . x$ 2 = and
for OSA, 0.03 s for DESS and 0.02 s for SESN, re- 1+ x 3 1+ x 3
spectively.
116 A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006)

T a b l e 3 – Bubble and dew point calculations for isobutene + methanol « methyl ter-butyl ether with n-butane as inert. Wilson
equation and ideal gas. xz (0.3, 0.3, 0.4)

Dew point Bubble point

T / °C
p / atm x$a p / atm x$ b

50 1.950 (0.43001, 0.45659, 0.11341) 3.339 (0.12692, 0.12009, 0.75299)

70 3.834 (0.41602, 0.44792, 0.13606) 5.713 (0.14796, 0.14425, 0.70779)

90 7.006 (0.40177, 0.43648, 0.16176) 9.302 (0.16826, 0.17076, 0.66099)

x4 sociation of acetic acid in the vapor phase, using a

x$ 4 = = 1- x$ 1 - x$ 2 , where the MTBE is the correction factor as suggested by Maier et al.29 The
1+ x 3
reference component. Again, the Wilson and correction factor zi for the association is given by
Antoine equations are used for the calculation of 1+ (1+ 4k1 p1sat )1/ 2
thermodynamic properties. Maier et al.29 report the z1 = (22)
parameters for both equations. The reaction equi- 1+ [1+ 4 k p x 1 (2 - x 1 )]1/ 2
librium constant is determined from DG 0rxs / R =
and
–4205.05 + 10.0982 T – 0.2667 T ln T, where T is in
K. For SESN and DESS methods, we use x$ 1 , x$ 2 and 2{1- x 1 + [1+ 4k P x 1 (2 - x 1 )]1/ 2 }
T as unknowns while in the optimization approach, zi =
the objective function is minimized with respect to (2 - x 1 ){1+ [1+ 4 k P x 1 (2 - x 1 )]1/ 2 }
x$ 1 , x$ 2 , x$ 4 and T. We set a feasible domain of (0, 1)
for all the transformed mole fractions and (0.0001, for i = 2, 3, 4 (23)
20) bar for pressure. The initials values are ran-
where p isat is the saturation pressure of pure com-
domly generated in these domains. For OSA
method, at each evaluation of the objective func- ponent i and the dimerization constant is calculated
tion, the transformed mole fractions are normalized from log10 k = log10 k1 = –12.5454 + 3166/T with
to unity. We have analyzed an initial global compo- T in Kelvin and k and k1 in Pa–1. Maier et al.29
sition of x$ F (0.3, 0.3, 0.4). Tab 3 shows the results gives the Antoine and NRTL equation parameters.
of bubble and dew point calculations at different The reaction equilibrium constant is independent of
temperatures and Tab 2 shows the performance of temperature, Keq = 8.7, and water is the refe-
the three equation-solving methods. For this case, rence component. The transformed mole frac-
x1 + x 4 x2 + x4
DESS shows 100 % reliability for the location of tions are given by x$ 1 = , x$ 2 = and
bubble and dew points at all tested temperatures, 1+ x 4 1+ x 4
while OSA method fails only one time in all calcu- x3 - x4
lations performed. Again, SESN shows several fail- x$ 3 = = 1- x$ 1 - x$ 2 . We use x$ 1 , x$ 2 and T as
1+ x 4
ures in the location of reactive saturation condi-
unknowns for all methods. The initial values are
tions. For dew point calculations, the performance
randomly generated using the feasible domain (0,
of SESN method is very poor. OSA method shows
1) for transformed mole fractions and (50, 200) °C
the highest computation time (22 s) while SESN
for temperature. Bubble and dew point calculations
and DESS showed a computation time of 0.08 and
are performed for a feed of x$ F (0.5, 0.3, 0.2) at
0.11 s, respectively. different pressures (see Tab 4 and Tab 2) shows
the numerical performance of all tested methods.
Example 3. DESS and OSA methods are very reliable to find
acetic acid + isopropanol « isopropyl acetate + water the bubble and dew point conditions and they did
not show any failures in all the calculations per-
Third example is the formation of isopropyl ac- formed. Again, SESN showed the worst perfor-
etate through esterification reaction of isopropanol mance and its numerical behavior is highly de-
and acetic acid: acetic acid (1) + isopropanol (2) « pendent on initial guesses. The mean number of
isopropyl acetate (3) + water (4). Harding and $
K-value evaluations is given by: SESN < DESS <<
Floudas30 and Maier et al.29 have studied this sys- OSA, while the computation time maintains around
tem. For the liquid phase, we use activity coeffi- of 17 s for OSA, 0.2 s for DESS and 0.03 s for
cients from the NRTL equation and consider the as- SESN.
A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006) 117

T a b l e 4 – Bubble and dew point calculations for acetic acid + isopropanol « isopropyl acetate + water. NRTL equation and
ideal gas, Keq = 8.7 and x$F (0.5, 0.3, 0.2).

Dew point Bubble point

p / atm
T / °C x$a T / °C x$ b
1 94.35 (0.70299, 0.1786, 0.11842) 85.55 (0.3791, 0.41569, 0.20522)
5 149.683 (0.63046, 0.14396, 0.22558) 138.272 (0.42525, 0.4785, 0.09625)
10 179.781 (0.59782, 0.13138, 0.27081) 166.451 (0.44323, 0.50417, 0.05260)

Example 4. computation time (3.3 seconds) and the highest

$
mean number of K-values evaluations. On the other
A1 + A2 « A3 and 2A3 « A4 + A2
hand, several failures are reported for the DESS and
Our last example is a hypothetical quaternary SESN methods. For non-reactive systems, Michel-
system that follows the reactions: A1 + A2 « A3 and sen31 has indicated that bubble and dew point calcu-
2A3 « A4 + A2. Ung and Doherty9 have studied the lations using hybrid models rarely present conver-
phase equilibrium behavior of this mixture. We gence problems. However, our results show that
have considered ideal behavior for both liquid and this doesn’t apply for phase equilibrium calcula-
vapor phases. The Antoine equation is used to cal- tions in reactive systems. Even for simple reacting
culate the vapor pressures of pure components with systems, SESN and DESS methods present conver-
the parameters reported by Ung and Doherty.9 The gence problems and depend on the initial guesses.
transformed mole fractions are defined using A3 and The OSA method is very reliable to find the bubble
A4 as reference components, so they are given by and dew points and its performance is almost inde-
x 1 + x 3 + 2x 4 x2 + x3 + x4 pendent of initial values. However, we can not offer
x$ 1 = and x$ 2 = = 1- x$ 1 . a theoretical guaranty for the location of bubble and
1+ x 3 + 2x 4 1+ x 3 + 2x 4
dew points using this optimization approach.
The chemical equilibrium constants are cal-
æ 7368ö Finally, we have tested the performance of these
culated using K 1eq = expç-2257 . + ÷ and equation-solving methods with other reactive sys-
è T ø tems reported by Okasinski and Doherty,28 Harding
æ 68441 . 1391790ö and Floudas30 and Maier et al.;29 and the results, not
K eq2
= expç-7.0265 + - ÷, where T
è T T2 ø reported in this paper, indicated that OSA and DESS
is given in Kelvin. We use x$ 1 and temperature as methods are reliable strategies for the calculation of
unknowns and their feasible domains are (0, 1) and dew and bubble points in multireactive mixtures.
(273.15, 400) K, respectively. Random initial val-
ues are generated inside these domains. Bubble and
dew point calculations are performed for a feed Conclusions
x$ F (0.5, 0.5) at several pressures. The results of
phase equilibrium calculations appear in Tab 5 and New conditions for bubble and dew point calcu-
performance of all tested methods are reported in lations in multireactive mixtures have been derived
Tab 2. Only, the OSA method showed 100 % reli- in terms of the variable transformation theory of Ung
ability to locate the bubble and dew points in this and Doherty.13 We have tested and compared the nu-
reactive system, however, it showed the highest merical performance of three equation-solving meth-

T a b l e 5 – Bubble and dew point calculations for a hypothetical quaternary mixture A1+A2 « A3 and 2A3 « A4+A2. Ideal solu-
tion and ideal gas. x$F (0.5, 0.5)

Dew point Bubble point

p / atm
T/K x$a T/K x$ b

1 355.488 (0.02678, 0.97322) 306.95 (0.96045, 0.03955)

4 395.265 (0.0452, 0.9548) 328.17 (0.97759, 0.02241)

6 409.034 (0.05399, 0.94601) 337.393 (0.97911, 0.02089)

8 419.525 (0.06133, 0.93867) 345.829 (0.97829, 0.02171)

 118 A. BONILLA-PETRICIOLET et al., Bubble and Dew Point Calculations in …, Chem. Biochem. Eng. Q. 20 (2) 111–118 (2006)

ods in the calculation of bubble and dew points us- Index

ing the new equilibrium conditions. Our results indi- F – feed composition
cate that an optimization approach using Simulated ref – reference component
Annealing is very reliable for this kind of phase F – phase
equilibrium calculations in multireactive systems.
However, it shows the highest computation time. In
References
the other hand, the equation-decoupling method with
1. Pérez-Cisneros, E. S., Gani, R., Michelsen, M. L., Chem.
successive substitution also presents a good numeri- Eng. Sci. 52 (1997) 527.
cal performance and is efficient. It is clear that the si- 2. Seider, W. D., Widagdo, S., Fluid Phase Equilib 123 (1996) 283.
multaneous equation solving approach, using New- 3. Castillo, J., Grossmann, I. E., Comp. Chem. Eng. 5 (1981) 99.
ton method, is not suitable for bubble and dew point 4. Castier, M., Rasmussen, P., Fredenslund, A., Chem. Eng.
calculations in reactive systems because it can di- Sci. 44 (1989) 237.
verge very frequently and depends significantly on 5. Xiao, W., Zhu, K., Yuan, W., Chien, H. H., An algorithm
initial values. for simultaneous chemical and phase equilibrium calcula-
tion. AIChE J. 35 (1989) 1813.
Based on reliability, we consider that the opti- 6. Greiner, H., An efficient implementation of newton’s
mization approach using Simulated Annealing is method for complex nonideal chemical equilibria. Comp.
the most suitable method and preferable choice for Chem. Eng. 15 (1991) 115.
these equilibrium calculations in reactive systems, 7. Smith, J. V., Missen, R. W., Smith, W. R., General
due to its acceptable convergence speed, and that optimality criteria for multiphase multireaction chemical
equilibrium. AIChE J. 39 (1993) 707.
initial guesses are not required. We will extend this 8. Ung, S., Doherty, M. F., AIChE J. 41 (1995) 2383.
procedure for performing these equilibrium calcula- 9. Ung, S., Doherty, M. F., Chem. Eng. Sci. 50 (1995) 23.
tions using equations of state. 10. McDonald, C. M., Floudas, C. A., Glopeq: a new compu-
tational tool for the phase and chemical equilibrium prob-
lem. Comp. Chem. Eng. 21 (1997) 1.
ACKNOWLEDGEMENTS 11. Stateva, R. P., Wakeham, W. A., Ind. Eng. Chem. Res. 36
(1997) 5474.
Authors acknowledge the financial support
12. Wasylkiewicz, S. K., Ung, S., Fluid Phase Equilib 175
from CONACYT, Instituto Tecnológico de Aguasca- (2000) 253.
lientes and PROMEP. 13. Ung, S., Doherty M. F., Chem. Eng. Sci. 50 (1995) 3201.
14. Eckert, E., Kubicek, M., Comp. Chem. Eng. 19 (1995) 493.
15. Walas, S. M., Phase equilibria in chemical engineering.
Notation Butterworth Publishers, USA, 1985.
16. Teh, Y. S., Rangaiah, G. P., Trans IChemE 80 (2002) 745.
a – activity, mol dm–3
17. Henderson, N., Freitas, L., Platt, G. M., AIChE J. 50
n – number of components (2004) 1300.
Keq – chemical equilibrium constant 18. Kirkpatrick, S., Gelatt, C., Vecci, M., Science 220 (1983) 671.
K – transformed phase equilibrium constant 19. Pan, H., Firoozabadi, A., Complex multiphase equilibria
calculations by direct minimization of Gibbs free energy
N – iteration number by use of simulated annealing. SPE Reservoir Eva. Eng.
p – pressure given in atm, 1 atm = 101 325 Pa Feb (1998) 36.
nR – reaction number 20. Zhu, Y., Xu, Z., Fluid Phase Equilib 154 (1999) 55.
21. Henderson, N., Oliveira, J. R., Ind. Eng. Chem. Res. 40
R – universal gas constant, J mol–1 K–1 (2001) 6028.
T – temperature given in °C or K 22. Gomes, J., Henderson, N., Rocha, M., Macromol. Theory
v ir – stoichiometric number of component i Simul. 10 (2001) 816.
23. Rangaiah, G. P., Fluid Phase Equilib 187-188 (2001) 83.
x$ – transformed mole fraction, 1 24. Bonilla-Petriciolet, A., Calculation of phase equilibrium in
x – mole fraction, 1 reactive and nonreactive systems. Ph. D. Dissertation, In-
x$ F – transformed feed composition, 1 stituto Tecnológico de Celaya, Guanajuato, Mexico, 2005.
25. Corana, A., Marchesi, M., Martini, C., Ridella, S., ACM
xF – feed composition, 1 T. Math. Software 13 (1987) 262.
x$ F – transformed phase mole fraction, 1 26. Metropolis, N., Rosenbluth, A., Rosenbluth, M., Teller, A.,
VM – vector length Teller E., J. Chem. Phys. 21 (1953) 1087.
27. Goffe, B., Ferrier, G., Rogers, J., Global optimization of
statistical functions with simulated annealing. J. Econo-
metrics 60 (1994) 65.
Greek letters
28. Okasinski, M. J., Doherty, M. F., AIChE J. 43 (1997) 2227.
a, b – phase at equilibrium 29. Maier R. W., Brennecke J. F., Stadtherr M. A., Comp.
Chem. Eng. 24 (2000) 1851.
m – chemical potential, J mol–1
30. Harding, S. T., Floudas, C. A., Ind. Eng. Chem. Res. 39
g – activity coefficient, 1 (2000) 1576.
j – fugacity coefficient, 1 31. Michelsen, M. L., Comp. Chem. Eng. 17 (1993) 431.

View publication stats