ARTICLE

pubs.acs.org/IECR

Group Contribution-Based Method for Determination of Solubility

Parameter of Nonelectrolyte Organic Compounds
Farhad Gharagheizi,† Ali Eslamimanesh,‡ Amir H. Mohammadi,*,‡,§ and Dominique Richon‡
†
Saman Energy Giti Co., 3331619636 Tehran, Iran
‡
MINES ParisTech, CEP/TEP
Centre E nergetique et Procedes, 35 Rue Saint Honore, 77305 Fontainebleau, France
§
Thermodynamics Research Unit, School of Chemical Engineering, University of KwaZulu-Natal, Howard College Campus,
King George V Avenue, Durban 4041, South Africa
bS Supporting Information
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

ABSTRACT: The determination of the solubility parameter of organic compounds has been of much significance in the chemical
industry. In this study, we propose a predictive method based on the combination of the Group Contribution strategy
Downloaded via TU EINDHOVEN on January 22, 2019 at 16:45:45 (UTC).

with the Artificial Neural Network to calculate/estimate the solubility parameter values of about 1620 nonelectrolyte organic
compounds at 298.15 K and atmospheric pressure. The chemical functional groups are obtained for various compounds categorized
in 81 different chemical families. The final results indicate the following statistical parameters of the presented method: average
relative deviation (ARD %) of the determined properties from existing experimental values of 1.5% and a squared correlation
coefficient of 0.985. It is finally inferred that the developed model is more accurate and predictive than our previously proposed
models based on the Quantitative Structure
Property Relationship algorithm, which yielded 4.6, 3.4, and 3.1 ARD % from
experimental values.

1. INTRODUCTION for defining the physical meaning of the solubility parameter as

Application of the solubility parameter concept in different follows:1,18
chemical processes has been investigated since 1930s, when  
∂P
Scatchard1 defined a physicochemical parameter standing for a Pi ¼ T
P ¼ δ2 ð2Þ
solvent's affinity to dissolve a particular solute.2 Hildebrand and ∂T v
Scott,3 and Hansen4,5 were the next researchers who improved where Pi stands for the internal pressure and T is temperature.
the definition of the solubility parameter along with its various The interactions between the solvents and the solutes, origi-
applications. So far, it has been demonstrated that the solubility nated from their electron pairs, donor
acceptors, and hydrogen
parameter can be employed as a prominent property of a bonding interactions, are not considered in the preceding
compound not only in coating and paint technologies, complex equations. In other words, the aforementioned Hildebrand’s
extraction operations, and polymer processes, and so forth2,6 parameter does not account for these interactions and considers
but also in many of the developed thermodynamic models for only one part of the molecular forces (dispersion).1,6 Conse-
prediction of the amounts/conditions of precipitations/ quently, the concept of the Hildebrand solubility parameter is
depositions of heavy petroleum fractions such as asphaltene normally applied for systems including weakly interacting spe-
and wax. Many of these models have been generally developed cies. As a result, Hildebrand’s theory was modified by several
based on the regular solution theory,7 which is based on the researchers, from various fields, to define the two-component
difference between the solubility parameter of the solute solubility parameter as follows:19
25
(asphaltene/wax) and related solvent (maltene/oil).8
17
Generally, the solubility parameter can be evaluated by the δ ¼ ðδλ 2 þ δτ 2 Þ1=2 ð3Þ
following equation:1
5
 1=2  1=2  1=2 where subscripts λ and τ denote nonpolar and polar solubility
ΔEv ΔUvap ΔHvap
RT parameters, respectively.
δ¼ ¼ ¼ ð1Þ
v v v As a very fruitful modification, Hansen5 proposed the three-
component (Hansen) solubility parameter considering the ef-
where δ stands for the Hildebrand one-component solubility fects of all of the cohesive bonds including the atomic dispersion
parameter, ΔEv represents the cohesive energy, which is introduced forces, the molecular permanent dipole-permanent dipole forces,
as the energy required for separating a molecule from its surrounded
neighbors,1
5 v is the molar volume, ΔUvap denotes the energy Received: May 9, 2011
change upon isothermal vaporization of the saturated liquid to the Accepted: August 2, 2011
ideal-gas state (energy of a complete vaporization),17 and ΔHvap is Revised: July 28, 2011
the enthalpy of vaporization. Internal pressure can also be applied Published: August 02, 2011

r 2011 American Chemical Society 10344 dx.doi.org/10.1021/ie201002e | Ind. Eng. Chem. Res. 2011, 50, 10344–10349
Industrial & Engineering Chemistry Research ARTICLE

and the molecular hydrogen-bonding on the solubility parameter 2.2. Determination of New Functional Groups. Having
value as follows:1,2,17 defined the database, the chemical structures of all of the studied
compounds have been analyzed in great detail using an algorithm
δHSP ¼ ðδD 2 þ δP 2 þ δH 2 Þ1=2 ð4Þ comparing the chemical groups to define the most efficient
contributions for evaluation of the solubility parameter. As a
where the subscripts D, P, and H denote the dispersion, polar, result, a new collection of 176 functional groups have been found
and hydrogen-bonding effects, respectively, and the subscript to be more efficient for the representation/prediction of the
HSP denotes the total Hansen solubility parameter. The values of corresponding parameter. These functional groups are more
one-component and the total three-component solubility para- general than those of first-order, second-order, or third-order
meters would be almost the same for the substances with groups used in conventional group contribution methods. The
nonpolar, and non-hydrogen-bonding effects such as the light functional groups used in this study are presented as Supporting
hydrocarbons. Information. Moreover, the table of their numbers of occur-
The determination of the solubility parameter has been rences in the investigated compounds is presented as Supporting
therefore of critical importance for the industry.1
39 A detailed Information.
review of the corresponding methods for its evaluation can be 2.3. Optimization of Group Contributions. The first calcula-
found elsewhere.17 This work is a continuation of the series of tion step is to find a relationship between the chemical functional
our efforts to develop predictive tools for the determination of groups and the desired physical property.41
82 The traditional
the physicochemical properties of nonelectrolyte organic com- and perhaps the easiest method for this purpose is the assump-
pounds using different approaches. In a previous communica- tion of the existence of a multilinear relationship between these
tion,17 our group presented three reliable models based on groups and the property (here the solubility parameter).41 This
Quantitative Structure
Property Relationship (QSPR) to repre- technique is a similar method to that used in the most of classical
sent/predict the one-component solubility parameter of none- group contribution methods.41 Several calculations show that
lectrolyte organic compounds reported in the DIPPR 801 application of the aforementioned methodology for the current
database.40 The results showed that the later models were reliable problem does not contribute to accurate results within the range
and comprehensive although developing such molecular-based of the deviations from experimental values40 that we are inter-
models may not contain easy computational procedure. In this ested in. Consequently, a nonlinear mathematical method such
work, a new approach based on the combination of the Group as ANN is preferred and investigated here. Using the Artificial
Contribution (GC) method with Artificial Neural Network Neural Network toolbox of the MATLAB software (Mathworks
(ANN) is presented for this purpose. One of the main char- Inc.), a three layer Feed Forward Artificial Neural Network
acteristics of the GC method is that this algorithm divides a (FFANN) has been developed for the problem.
molecule into small parts (generally named as “segments”). Each Because we face a wide range of solubility parameter values
of these segments is considered as a functional group and has a for different compounds, these values have been normalized
contribution to the physicochemical properties of the specified between
1 and +1 to prevent truncation errors.41 This can
molecule. Finally, the value of the property is defined through be performed using maximum and minimum numbers of each
calculating the summation of the contributions of all functional functional group in each compound for input data and using
groups in a molecule. maximum and minimum values of solubility parameter for output
Furthermore, the Artificial Neural Networks have been applied parameters.41 In addition, this procedure, which is done in the
to various scientific and engineering applications,41
82 for example, optimization process, is performed to obtain the parameters of the
calculations/estimations of the physical properties of different pure Neural Networks (weights and bias), and it has no effects on the
compounds41
45 and phase behavior predictions of complex semi- model results. Later, these values are again changed to the original
clathrate hydrate systems.60 The theoretical explanations about solubility parameter values, which are finally used as the inputs and
Neural Networks have been well-established elsewhere.83 As a reported as outputs of the developed model.41 In the second step,
consequence, a combination of the GC and the ANN methods the database is divided into three subdata sets including the
normally leads to obtain accurate predictive tools for the evaluation “Training” set, the “Optimization” set, and the “Test” set. In this
of the desired properties of organic compounds. However, ANN is a study, the “Training” set is used to generate the ANN structure, the
mathematical tool that users must be very careful to apply its “Optimization” set is applied for optimization of the model, and the
consequent results within the frame of the hypotheses and within “Test (prediction)” set is used to investigate the prediction cap-
the field of the data that allowed determination of the parameters ability and validity of the proposed model. The division of database
(any extrapolation may not be recommended). into three subdata sets is normally performed randomly. For this
purpose, about 80%, 10%, and 10% of the main data set are
2. EXPERIMENTAL DATA AND MATHEMATICAL randomly selected for the “Training” set (about 1296 solubility
METHODS parameter data), the “Optimization” set (162 solubility parameter
data), and the “Test” set (162 solubility parameter data). The effect
2.1. Experimental Data. In this study, we have used the DIPPR of the percent allocation of the three subdata sets from the database
801 database,40 which is one of the most reliable sources of physical on the accuracy of the final model has been studied elsewhere.76 In
property data for pure compounds, based on more than 23000 distribution of the data through the three subdata sets, we generally
scientific sources. The solubility parameter values of 1620 none- perform many distributions to avoid the local accumulations of the
lectrolyte organic species from various chemical families (81 data in the feasible region of the problem. As a result, the acceptable
families) at 298.15 K and atmospheric pressure have been treated distribution is the one with homogeneous accumulations of the data
for the calculation procedure. All of the data points have been on the domain of the three subdata sets.
evaluated by the DIPPR 801 project40 for the investigated There are generally two weight matrices and two bias vectors
compounds. in a three layer FFANN: W1 and W2, b1 and b2. These parameters
10345 dx.doi.org/10.1021/ie201002e |Ind. Eng. Chem. Res. 2011, 50, 10344–10349
Industrial & Engineering Chemistry Research ARTICLE

should be evaluated by minimization of an objective function. The

objective function here is the summation of squares of errors
between the outputs of the ANN (represented/predicted properties)
and the target values (experimental solubility parameters). This
minimization is performed by the Levenberg
Marquardt (LM)83
optimization strategy. There are also more accurate optimization
methods other than this algorithm; however, they need much
more convergence time.41
48 The more accurate optimization, the
more time is needed for the algorithm to converge to the global
optimum. The LM83 is the most-widely used optimization method
in these kinds of problems.41
In most cases, the number of neurons in the hidden layer (n) is
fixed. Therefore, the objective is to produce the ANN model, which
is capable of predicting the target values as accurately as possible.
This step is repeated until the best ANN is obtained. Normally, in
three-layer FFANNs, it is more efficient that the number of neurons
in the hidden layer is optimized according to the accuracy of the
produced FFANN.29
61 Some factors should be taken into account
Figure 1. Comparison between the Cal (calculated)/Est (estimated)
in the determination of the optimum number of the neurons. By results using the developed model and experimental values40 of solubi-
increasing the number of neurons, the accuracy of the model, that lity parameters.
is, squared correlation coefficient (R2), is increased on the “Train-
ing set”. However, the accuracy of the model on the “Test set” is
decreased gradually, and the model may become unstable. Conse-
quently, the overall R2, which depends on the three subdata sets,
fluctuates during the changing of the numbers of neurons. The final
(overall) R2 value should be found through selecting the different
number of neurons for a specified problem.17

3. RESULTS AND DISCUSSION

An optimized GC-ANN model has been obtained applying
the preceding procedure for the determination of the desired
parameter. For this purpose, several 3FFANNs modules were
generated assuming numbers 1 through 50 for n (number of
neurons in hidden layer). The most accurate results (no over-
fitted and no under-fitted results) were observed at n = 10. It
should be noted that this value is not a global one, because the
optimization method used to train the ANN has great effects on
the obtained value.17,41 Therefore, the developed three-layer
FFANN has the structure of 176-10-1 (176 chemical groups are
regarded as the inputs of the algorithm).
A significant point that needs to be considered about the
number of the model parameters is that, in each compound, only Figure 2. Deviations of the obtained results vs the corresponding
a few functional groups are present simultaneously (with the experimental (Exp) solubility parameter values.40
maximum number of 27 groups in pimaric acid and isopimaric
acid). Therefore, for each compound, many of the model may not be used as the only criterion for a comparison between
parameters, which in total are 10  176 = 1760, are zero and the developed models because they are based on different
consequently, the developed model has few parameters for each network concepts with different mathematical characteristics,
compound, that is, between 0 to 270 parameters. The value of which have been well-established in our previous work.17
zero indicates that three of the investigated compounds do not The mat file (MATLAB file format) of the obtained GC-ANN
contain the determined functional groups in their structures by containing all the parameters of the model (weight matrices
the previous computational step. For these compounds, the and bias vectors) and the instructions for running the program
model results in the solubility parameter value of 18.19 are freely available upon request to the authors. Running the
(J/cm3)0.5, which has been calculated by the intercept of the provided software, any researcher/engineer is able to determine
transfer function of the ANN algorithm (Refer to the Supporting the solubility parameter of a particular substance quickly. The
Information file to observe the characteristics of all of the employed functional groups to develop the model have been
investigated organic compounds in this work). It should be reported as Supporting Information. The calculated/estimated
pointed out that the number of the model parameters for the solubility parameter values are shown in Figure 1 in comparison
previous developed models17 were 11, 131, and 13 for the linear- with the experimental values.40 Figure 2 indicates the deviations
QSPR, ANN-QSPR, and LSSVM-QSPR models, respectively. of the obtained values versus the experimental ones.40
However, the number of parameters for the two later nonlinear The statistical results obtained by the GC-ANN model are
QSPR models17 and the proposed GC-ANN model in this study reported in Table 1. Furthermore, the absolute average deviations
10346 dx.doi.org/10.1021/ie201002e |Ind. Eng. Chem. Res. 2011, 50, 10344–10349
Industrial & Engineering Chemistry Research ARTICLE

Table 1. Statistical Parameters of the Proposed GC-ANN the imperative need for the development of new experimental
Model techniques and apparatuses to generate more solubility parameter
data for particular industrial and theoretical purposes.
statistical parameter value

Training Set 4. CONCLUSION

R2 a 0.985
In this work, a group contribution-based model was presented
average relative deviation b 1.5% for the representation/prediction of the solubility parameter of
standard deviation error, c (J cm
3)0.5 3.86 nonelectrolyte organic compounds at 298.15 K and atmospheric
root mean square error 0.48 pressure. The model is the result of a combination of Feed
Nd 1296 Forward Artificial Neural Networks and Group Contributions.
Optimization Set
The required parameters of the model are the numbers of
2 occurrences of 176 functional groups in each investigated
R 0.991
molecule. It should be noted that most of these functional groups
average relative deviation 1.3% are not simultaneously available in a particular molecule. There-
standard deviation error, (J cm
3)0.5 3.95 fore, the computation of the required parameters from the
root mean square error 0.38 chemical structure of any molecule is simple. For developing
N 162 the model, the experimental solubility parameter values from the
Test Set
largest available data set40 containing 1620 pure organic com-
2 pounds from 81 different chemical families were applied. As a
R 0.991
consequence, a reliable and predictive tool was developed to
average relative deviation 1.6%
determine the solubility parameter values of many of organic
standard deviation error, (J cm
3)0.5 4.06 compounds, which are especially applied in chemical and petro-
root mean square error 0.55 leum industry. However, one point should not be omitted from
N 162 our conclusion: The model has a wide range of applicability but
Training + Optimization + Test Set the prediction capability of the model is restricted to the
compounds which are similar to the applied to develop the
R2 0.985
model. Application of the model for totally different compounds
average relative deviation 1.5%
than the investigated ones is conservative although it may be used
standard deviation error, (J cm
3)0.5 3.89 for a rough estimation of the solubility parameter of these kinds
root mean square error 0.48 of compounds.
N 1620
 N


R ¼ 1
∑i
2 Cal:ðiÞ=Est:ðiÞ

a
R2: Squared correlation coefficient, 2
2
Exp:ðiÞ =∑i Cal:ðiÞ=Est:ðiÞ
average Exp:ðiÞ g.
N b
%ARD = ’ ASSOCIATED CONTENT
(100)/(N
n)∑iN(|Cal.(i)/Est.(i)
Exp.(i)|)/(Exp.(i)), where n is the
number of the model parameters. c std = (1/N)∑iN((Cal.(i)/Est.(i)
bS Supporting Information. There are four files including
average(Cal.(i)/Est.(i)))2)1/2. the number of occurrences of the 176 functional groups in all of
d 1620 nonelectrolyte organic compounds in the main data set, the
Number of data points.
distributions of the data in three data sets, the obtained results in
the first Excel file, the functional groups used to develop the
of the results from experimental values40 for each 81 chemical model in the second Excel file, the average absolute deviations of
families are reported as Supporting Information. The results imply the obtained results from experimental values of various chemical
that the obtained ANN-GC model is more accurate and predictive families in the third Excel file, and the schematic structures of all
than our previously presented models17 based on the QSPR of the investigated compounds in the pdf format file. This
strategy. For better illustration of the organic compounds inves- material is available free of charge via the Internet at http://
tigated in this work, their chemical structures are sketched and pubs.acs.org.
presented as Supporting Information.
It should be noted that we have observed that there are 17 data
’ AUTHOR INFORMATION
points (from 1620 available data points) for which the presented
model results lead to more than 10% average relative deviations Corresponding Author
from experimental values.40 It seems that there is no relation *E-mail: [email protected]. Phone:
between these compounds' structures to show some weaknesses + (33) 1 64 69 49 70. Fax: + (33) 1 64 69 49 68.
in determining the solubility parameter values of related chemical
families. Therefore, it is probable that the solubility parameter values
for these compounds are not accurate or may be somehow ’ ACKNOWLEDGMENT
erroneous (or with high uncertainty) because of the existing diffi- Ali Eslamimanesh wishes to thank MINES ParisTech for
culties and possible errors in experimental measurements.84
87 providing him a Ph.D. scholarship.
The proposed method has been developed based on all of the
available data including the aforementioned 17 data points. The
final significant point is that we do not have access to any other ’ REFERENCES
solubility parameter database to check the capability of the pre- (1) Scatchard, G. Equilibria in Non-electrolyte solutions in relation
sented model for prediction of solubility parameters of completely to the vapor pressures and densities of the components. Chem. Rev.
nonsimilar compounds to those investigated. This fact again proves 1931, 8, 321–333.

10347 dx.doi.org/10.1021/ie201002e |Ind. Eng. Chem. Res. 2011, 50, 10344–10349

Industrial & Engineering Chemistry Research ARTICLE

(2) Eslamimanesh, A.; Esmaeilzadeh, F. Estimation of solubility (25) Lyckman, E. W.; Eckert, C. A.; Prausnitz, J. M. Generalized
parameter by the modified ER equation of state. Fluid Phase Equilib. liquid volumes and solubility parameters for regular solution application.
2010, 291, 141–150. Chem. Eng. Sci. 1965, 20, 703–706.
(3) Hildebrand, J. H.; Scott, R. L. The Solubility of Nonelectrolytes; (26) Fedors, R. F. A method for estimating both the solubility
Rheinhold Publishing Corporation: New York, 1950. parameter and molar volumes of liquids. Polym. Eng. Sci. 1974, 14,
(4) Hansen, C. M. The three dimensional solubility parameter - key 147–154.
to paint component affinities I. - Solvents, plasticizers, polymers, and (27) Lozada, C. M.; Del Rio, F. Calculation of solubility parameters
resins. J. Paint Technol. 1967, 39, 104–117. from an equation of state. Chem. Phys. Lett. 1977, 45, 130–1130.
(5) Hansen, C. M. The Universality of the Solubility Parameter. Ind. (28) Hansen, C. M.; Beerbower, A. Solubility parameters. In Kirk-
Eng. Chem. Pro. Res. Develop. 1969, 8, 2–11. Othmer Encyclopedia of Chemical Technology, Suppl. Vol., 2nd ed.;
(6) Gharagheizi, F.; Sattari, M.; Angaji, M. T. Effect of calculation Standen, A., Ed.; Interscience: New York, 1971.
method on values of Hansen solubility parameters of polymers. Polym. (29) Beerbower, A. Environmental Capability of Liquids, in Inter-
Bull. 2006, 57, 377–384. disciplinary Approach to Liquid Lubricant Technology. NASA Publica-
(7) Hildebrand, J. H. Solubility: XII. Regular Solutions. J. Am. Chem. tion SP-318; NASA: Washington, DC, 1973.
Soc. 1929, 51, 66–80. (30) Van Krevelen, D. W.; Hoftyzer, P. J. Properties of Polymers: Their
(8) Mousavi-Dehghani, S. A.; Mirzayi, B.; Mousavi, S. M. H.; Fasih, Estimation and Correlation with Chemical Structure, 2nd ed.; Elsevier:
M. An applied and efficient model for asphaltene precipitation in Amsterdam, The Netherlands, 1976.
production and miscible gas injection. Pet. Sci. Technol. 2010, (31) Allada, S. R. Solubility parameters of supercritical fluids. Ind.
28, 113–124. Eng. Chem. Prod. Res. Dev. 1984, 23, 344–348.
(9) Novosad, Z.; Costain, T. G. Experimental and modeling studies (32) Panayiotou, C. Solubility parameter revisited: An equation-of-
of asphaltene equilibria for a reservoir under CO2 injection. In Proceed- state approach for its estimation. Fluid Phase Equilib. 1997, 131, 21–35.
ings of the 65th Annual Technical Conference and Exhibition of the SPE, (33) Williams, L. L.; Rubin, J. B.; Edwards, H. W. Calculation of
SPE 20530, New Orleans, Los Angeles, United States, 1990. Hansen solubility parameter values for a range of pressure and tem-
(10) Nor-Azian, N.; Adewumi, M. A. Development of asphaltene perature conditions, including the supercritical fluid region. Ind. Eng.
phase equilibrium predictive model. In Proceedings of the Eastern Regional Chem. Res. 2004, 43, 4967–4972.
Conference and Exhibition of the SPE, SPE 26905, Nov., Richardson, (34) Bozdogan, A. E. A method for determination of thermody-
Texas, 1993. namic and solubility parameters of polymers from temperature and
(11) MacMillan, D. J.; Tackett, J. E.; Jessee, M. A.; Monger-McClure, molecular weight dependence of intrinsic viscosity. Polymer 2004,
T. G. A unified approach to asphaltene precipitation: Laboratory 45, 6415–6424.
measurement and modeling. In Proceedings of the SPE International (35) Utracki, L. A.; Simha, R. Statistical thermodynamics predictions
Symposium on Oilfield Chemistry, San Antonio, Texas, Feb., SPE 28990, of the solubility parameter. Polym. Int. 2004, 53, 279–286.
Richardson, Texas, 1995. (36) Stefanis, E.; Tsivintzelis, I.; Panayiotou, C. The partial solubility
(12) Yang, Z.; Ma, C. F.; Lin, S. X.; Yang, J. T.; Guo, T. M. Experimental parameters: An equation-of-state approach. Fluid Phase Equilib. 2006,
and modeling studies on the asphaltene precipitation in degassed and gas- 240, 144–154.
injected reservoir oils. Fluid Phase Equilib. 1999, 157, 143–158. (37) Zeng, Z.-Y.; Xu, Y.-Y.; Li, Y.-W. Calculation of solubility
(13) Alboudwarej, H.; Akbarzadeh, K.; Beck, J.; Svercek, W. Y.; parameter using Perturbed-Chain SAFT and Cubic-Plus-Association
Yarranton, H. W. Regular solution model of asphaltene precipitation equations of state. Ind. Eng. Chem. Res. 2008, 47, 9663–9669.
from bitumen. AIChE J. 2003, 11, 2948–2956. (38) Code, J. E.; Holder, A. J.; Eick, J. D. Direct and indirect
(14) Akbarzadeh, K.; Alboudwarej, H.; Svercek, H. Y.; Yarranton, quantum mechanical QSPR Hildebrand solubility parameter models.
H. W. A generalized regular solution model for the prediction of QSAR Comb. Sci. 2008, 27, 841–849.
asphaltene precipitation from n-alkane diluted heavy oils and bitumens. (39) Vargas, F. M.; Chapman, W. G. Application of the One-Third
Fluid Phase Equilib. 2005, 232, 159–170. rule in hydrocarbon and crude oil systems. Fluid Phase Equilib. 2010,
(15) Yarranton, H. W.; Masliyah, J. H. Molar mass distribution and 290, 103–108.
solubility modeling of asphaltenes. AIChE J. 1996, 42, 3533–3543. (40) Project 801, Evaluated Process Design Data, Public Release
(16) Mohammadi, A. H.; Richon, D. The Scott-Magat polymer Documentation, Design Institute for Physical Properties (DIPPR),
theory for determining onset of precipitation of dissolved asphaltene American Institute of Chemical Engineers (AIChE), 2006.
in the solvent + precipitant solution. Open Thermodyn. J. 2008, 2, 13–16. (41) Gharagheizi, F.; Eslamimanesh, A.; Mohammadi, A. H.;
(17) Gharagheizi, F.; Eslamimanesh, A.; Farjood, F.; Mohammadi, Richon, D. Use of artificial neural network-group contribution method
A. H.; Richon, D. Solubility parameter of non-electrolyte organic com- to determine surface tension of pure compounds. J. Chem. Eng. Data
pounds: Determination using quantitative structure
property relation- 2011, 56, 2587
2601.
ship strategy. Ind. Eng. Chem. Res. 2011, accepted manuscript, DOI: (42) Gharagheizi, F.; Sattari, M. Prediction of triple-point tempera-
10.1021/ie200962w. ture of pure components using their chemical structures. Ind. Eng. Chem.
(18) Van Arkel, A. E. Mutual solubility of liquids. Faraday Trans. Res. 2010, 49, 929–932.
1946, 42B, 81–84. (43) Gharagheizi, F.; Abbasi, R.; Tirandazi, B. Prediction of Henry’s
(19) Small, P. A. Some factors affecting the solubility of polymers. law constant of organic compounds in water from a new group-
J. Appl. Chem. 1953, 3, 71–80. contribution-based model. Ind. Eng. Chem. Res. 2010, 49, 10149–10152.
(20) Anderson, R. Polar organic solvents and aromatic hydrocar- (44) Eslamimanesh, A.; Gharagheizi, F.; Mohammadi, A. H.; Richon,
bons. Ph.D. Thesis, Dept. of Chemical Engineering, University of D. Artificial neural network modeling of solubility of supercritical carbon
California, Berkeley, CA, 1961. dioxide in 24 commonly used ionic liquids. Chem. Eng. Sci. 2011, 66,
(21) Blanks, R. F.; Prausnitz, J. M. Thermodynamics of polymer solubility 3039–3044.
in polar and nonpolar systems. Ind. Eng. Chem. Fundam. 1964, 3, 1–8. (45) Gharagheizi, F.; Eslamimanesh, A.; Mohammadi, A. H.; Richon,
(22) Pausnitz, J. M.; Anderson, R. Thermodynamics of solvent selec- D. Artificial neural network modeling of solubilities of 21 mostly-used
tivity in extractive distillation of hydrocarbons. AIChE J. 1961, 7, 96–101. industrial solid compounds in supercritical carbon dioxide. Ind. Eng. Chem.
(23) Weimer, R. F.; Prausnitz, J. M. Complex formation between Res. 2011, 50, 221–226.
carbon tetrachloride and aromatic hydrocarbons. J. Chem. Phys. 1965, (46) Gharagheizi, F.; Eslamimanesh, A.; Mohammadi, A. H.; Richon,
42, 3643–3644. D. Representation/prediction of solubilities of pure compounds in water
(24) Prausnitz, J. M.; Shair, F. H. A thermodynamic correlation of using artificial neural network
group contribution method. J. Chem. Eng.
gas solubilities. AIChE J. 1961, 7, 682–687. Data 2011, 56, 720–726.

10348 dx.doi.org/10.1021/ie201002e |Ind. Eng. Chem. Res. 2011, 50, 10344–10349

Industrial & Engineering Chemistry Research ARTICLE

(47) Gharagheizi, F.; Eslamimanesh, A.; Mohammadi, A. H.; Richon, (67) Sattari, M.; Gharagheizi, F. Prediction of molecular diffusivity of
D. QSPR approach for determination of parachor of non-electrolyte pure components into air: A QSPR approach. Chemosphere 2008,
organic compounds. Chem. Eng. Sci. 2011, 66, 2959–2967. 72, 1298–1302.
(48) Gharagheizi, F.; Eslamimanesh, A.; Mohammadi, A. H.; Richon, (68) Gharagheizi, F.; Alamdari, R. F.; Angaji, M. T. A new neural
D. Determination of parachor of various compounds using an artificial network-group contribution method for estimation of flash point. Energy
neural network - group contribution method. Ind. Eng. Chem. Res. 2011, Fuels 2008, 22, 1628–1635.
50, 5815
5823. (69) Gharagheizi, F.; Fazeli, A. Prediction of Watson characteriza-
(49) Chouai, A.; Laugier, S.; Richon, D. Modeling of thermody- tion factor of hydrocarbon compounds from their molecular properties.
namic properties using neural networks: Application to refrigerants. QSAR Comb. Sci. 2008, 27, 758–767.
Fluid Phase Equilib. 2002, 199, 53–62. (70) Gharagheizi, F.; Alamdari, R. F. A molecular-based model for
(50) Piazza, L.; Scalabrin, G.; Marchi, P.; Richon, D. Enhancement prediction of solubility of c60 fullerene in various solvents. Fullerenes,
of the extended corresponding states techniques for thermodynamic Nanotubes, Carbon, Nanostruct. 2008, 16, 40–57.
modelling. I. Pure fluids. Int. J. Refrig. 2006, 29, 1182–1194. (71) Gharagheizi, F. A new neural network Quantitative Structure-
(51) Scalabrin, G.; Marchi, P.; Bettio, L.; Richon, D. Enhancement of Property Relationship for prediction of θ (Lower Critical Solution
the extended corresponding states techniques for thermodynamic Temperature) of polymer solutions. e-Polym. 2007, 114, 1–5.
modelling. II. Mixtures. Int. J. Refrig. 2006, 29, 1195–1207. (72) Gharagheizi, F. QSPR studies for solubility parameter by means
(52) Chapoy, A.; Mohammadi, A. H.; Richon, D. Predicting the of genetic algorithm-based multivariate linear regression and generalized
hydrate stability zones of natural gases using Artificial Neural Networks. regression neural network. QSAR Comb. Sci. 2008, 27, 165–170.
Oil Gas Sci. Technol.
Rev. IFP 2007, 62, 701–706. (73) Gharagheizi, F. A chemical structure-based model for estima-
(53) Mohammadi, A. H.; Richon, D. Hydrate phase equilibria for tion of upper flammability limit of pure compounds. Energy Fuels 2010,
hydrogen + water and hydrogen + tetrahydrofuran + water systems: 27, 3867–3871.
Predictions of dissociation conditions using an artificial neural network (74) Vatani, A.; Mehrpooya, M.; Gharagheizi, F. Prediction of
algorithm. Chem. Eng. Sci. 2010, 65, 3352–3355. standard enthalpy of formation by a QSPR Model. Int. J. Mol. Sci. 2007,
(54) Mohammadi, A. H.; Richon, D. Estimating sulfur content of 8, 407–432.
hydrogen sulfide at elevated temperatures and pressures using an (75) Mehrpooya, M.; Gharagheizi, F. A Molecular approach for
artificial neural network algorithm. Ind. Eng. Chem. Res. 2008, 47, prediction of sulfur compounds solubility parameters, phosphorus sulfur
8499–8504. and silicon and related elements. Phosphorus, Sulfur Silicon Relat. Elem.
(55) Mohammadi, A. H.; Richon, D. A mathematical model based 2010, 185, 204–210.
on artificial neural network technique for estimating liquid water
(76) Gharagheizi, F. QSPR analysis for intrinsic viscosity of polymer
hydrate equilibrium of water
hydrocarbon System. Ind. Eng. Chem. solutions by means of GA-MLR and RBFNN. Comput. Mater. Sci. 2007,
Res. 2008, 47, 4966–4970. 40, 159.
(56) Mohammadi, A. H.; Afzal, W.; Richon, D. Determination of (77) Gharagheizi, F. An accurate model for prediction of autoigni-
critical properties and acentric factors of petroleum fractions using tion temperature of pure compounds. J. Hazard. Mater. 2011, 189,
artificial neural networks. Ind. Eng. Chem. Res. 2008, 47, 3225–3232. 211–221.
(57) Mohammadi, A. H.; Richon, D. Use of artificial neural networks (78) Gharagheizi, F.; Sattari, M.; Tirandazi, B. Prediction of lattice
for estimating water content of natural gases. Ind. Eng. Chem. Res. 2007, crystal energy using enthalpy of sublimation: A group contribution-
46, 1431–1438. based model. Ind. Eng. Chem. Res. 2011, 50, 2482–2486.
(58) Mohammadi, A. H.; Martínez-Lopez, J. F.; Richon, D. Deter- (79) Gharagheizi, F.; Eslamimanesh, E.; Mohammadi, A. H.; Richon,
mining phase diagrams of tetrahydrofuran+methane, carbon dioxide or D. Representation/prediction of molecular diffusivity of pure com-
nitrogen clathrate hydrates using an artificial neural network algorithm. pounds in water at infinite dilution using artificial neural network-group
Chem. Eng. Sci. 2010, 65, 6059–6063. contribution method. J. Chem. Eng. Data 2011, 56, 1741
1750.
(59) Mehrpooya, M.; Mohammadi, A. H.; Richon, D. Extension of (80) Gharagheizi, F.; Babaie, O.; Mazdeyasna, S. Prediction of vapor-
an artificial neural network algorithm for estimating sulfur content of ization enthalpy of pure compounds using a group contribution-based
sour gases at elevated temperatures and pressures. Ind. Eng. Chem. Res. method. Ind. Eng. Chem. Res. 2011, 50, 6503
6507.
2010, 49, 439–442. (81) Gharagheizi, F.; Mirkhani, S. A.; Mahyari, A. R. T. Prediction of
(60) Mohammadi, A. H.; Belandria, V.; Richon, D. Use of an artificial standard enthalpy of combustion of pure compounds using a very
neural network algorithm to predict hydrate dissociation conditions for accurate group contribution-based method. Energy Fuels 2011, 25,
hydrogen + water and hydrogen + tetra-n-butyl ammonium bromide + 2651
2654.
water systems. Chem. Eng. Sci. 2010, 65, 4302–4305. (82) Gharagheizi, F.; Salehi, G. R. Prediction of enthalpy of fusion of
(61) Gharagheizi, F. A new group contribution-based method for pure compounds using an artificial neural network-group contribution
estimation of lower flammability limit of pure compounds. J. Hazard. method Thermochim. Acta 2011, 521, 37
40.
Mater. 2009, 170, 595–604. (83) Hagan, M.; Demuth, H. B.; Beale, M. H. Neural Network Design;
(62) Gharagheizi, F. New Neural Network Group Contribution International Thomson Publishing Co.: Salt Lake City, UT, 2002.
model for estimation of lower flammability limit temperature of pure (84) Mohammadi, A. H.; Eslamimanesh, A.; Richon, D. Wax solu-
compounds. Ind. Eng. Chem. Res. 2009, 48, 7406–7416. bility in gaseous system: Thermodynamic consistency test of experi-
(63) Gharagheizi, F.; Sattari, M. Estimation of molecular diffusivity mental data. Ind. Eng. Chem. Res. 2011, 50, 4731–4740.
of pure chemicals in water: A quantitative structure-property relation- (85) Eslamimanesh, A.; Yazdizadeh, M.; Mohammadi, A. H.; Richon,
ship study. SAR QSAR Environ. Res. 2009, 20, 267–285. D. Experimental data assessment test for diamondoids solubility in gaseous
(64) Gharagheizi, F. Prediction of standard enthalpy of formation of system. J. Chem. Eng. Data 2011, 56, 2655
2659.
pure compounds using molecular structure. Aust. J. Chem. 2009, 62, (86) Eslamimanesh, A.; Mohammadi, A. H.; Richon, D. Thermo-
376–381. dynamic consistency test for experimental data of water content of
(65) Gharagheizi, F.; Tirandazi, B.; Barzin, R. Estimation of aniline methane. AIChE J. 2011, 57, 2566
2573.
point temperature of pure hydrocarbons: A Quantitative Structure- (87) Coquelet, C.; Galicia-Luna, L. A.; Mohammadi, A. H.; Richon,
Property Relationship approach. Ind. Eng. Chem. Res. 2009, 48, D. The essential importance of experimental research and the use of
1678–1682. experimental thermodynamics to the benefit of industry. Fluid Phase
(66) Gharagheizi, F.; Mehrpooya, M. Prediction of some important Equilib. 2010, 296, 2–3.
physical properties of sulfur compounds using QSPR models. Mol.
Diversity 2008, 12, 143–155.

10349 dx.doi.org/10.1021/ie201002e |Ind. Eng. Chem. Res. 2011, 50, 10344–10349