Accepted Manuscript

Solubility of grape seed oil in supercritical CO2: Experiments and modeling

Kurabachew Simon Duba, Luca Fiori

PII: S0021-9614(16)30038-6
DOI: http://dx.doi.org/10.1016/j.jct.2016.04.010
Reference: YJCHT 4625

To appear in: J. Chem. Thermodynamics

Received Date: 19 October 2015

Revised Date: 7 April 2016
Accepted Date: 15 April 2016

Please cite this article as: K.S. Duba, L. Fiori, Solubility of grape seed oil in supercritical CO2: Experiments and
modeling, J. Chem. Thermodynamics (2016), doi: http://dx.doi.org/10.1016/j.jct.2016.04.010

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Solubility of grape seed oil in supercritical CO2: Experiments
and modeling
Kurabachew Simon Duba, Luca Fiori*

Department of Civil, Environmental and Mechanical Engineering, University of Trento, via

Mesiano 77, 38123 Trento (TN), Italy

*
Email: luca.fiori@unitn.it Tel.: +39 0461 282692 Fax: +39 0461 282672

ABSTRACT

The solubility of grape (Vitis vinifera L.) seed oil in supercritical CO2 was measured in the
temperature range 313-343 K and pressure range 20-50 MPa using the dynamic technique. Several
data and global trends were reported. The results show that, at constant temperature, the solubility
increases with the increase in pressure, while the effect of the temperature is different for low and
high pressure. The experimental data were modeled by eight density-based models and a
thermodynamic model based on the Peng-Robinson equation of state. By best fitting procedures, the
“free parameters” of the various models were calculated: in general, all the tested models have
proved to be able to predict the solubility of grape seed oil in supercritical CO2. Differences in
model capabilities have been discussed based on the main characteristics of the various models,
evidencing their distinct and common features. The predictive capability of the thermodynamic
model was comparable to that of the density-based models.

Keywords: Grape seed oil, Solubility, Supercritical CO2 extraction, Density-based models,
Thermodynamic model

1. Introduction

Grape seeds are a by-product of the wine-making process; they contain between 4 and17% oil [1,2].
Grape seed oil consists mainly of triacylglycerols, with minor amounts of 1,2 diacylglycerols and
oxidized lipids [2]. The major fatty acids of grape seed oil are linoleic acid (C18:2 ≈70%), oleic
acid (C18:1 ≈15%), palmitic acid (C16:0 ≈7%) and stearic acid (C18:0 ≈3%) [1–3]. Grape seed oil
is a noble source of vitamin E, with substantial amounts of tocopherols and tocotrienols [1,2], and
exhibits high antioxidant activities which make it increasingly attractive in culinary, cosmetics,
pharmaceutical and medical applications [1,4,5].
Supercritical fluids (especially CO2) are receiving a central attention as future industrial solvents,
particularly in the field of high value-added products and applications. Supercritical fluids possess
gas-like viscosity and diffusivity, and liquid-like density and solvating power [6]. When used as
solvents, they can be easily separated from the extracted product by simple depressurization.
Besides, the thermodynamic properties of supercritical fluids can be easily tuned by adjusting the
operating conditions. Supercritical CO2 (SC-CO2), in addition to the advantages mentioned above,
presents other obvious strengths: non-toxicity, non-flammability, and low-cost.

The solvent power of SC-CO2 depends essentially on the nature of the solute and the system
pressure and temperature. An effective design and scale-up of supercritical fluids processes, such as
supercritical fluid extraction and fractionation, require the knowledge of the thermodynamics of the
system, i.e. the phase equilibria [7]. Thus, the determination of the solubility of a substance in the
supercritical phase is the first step in the development and evaluation of any supercritical process,
and it is also mandatory for establishing the optimal operating conditions [8–11].

The experimental determination of the solubility in supercritical fluids is generally accomplished

through two approaches in the literature, the so-called static and dynamic methods. In the static
method, solute and solvent are placed within a fixed volume vessel, which is stirred mechanically or
by recirculating the vapor phase until equilibrium is established. In the dynamic method, a
continuous apparatus is used to contact the two phases and the composition of the stream leaving
the apparatus is determined after expansion and separation of the solute from the supercritical
solvent [8,12]. Significant research works were published in the past two decades focused on the
determination of solubility of various organic compounds in SC-CO2, such as drugs [10,13,14],
seed oils [15–17], pollutants [18], dyes [19–21], food colorants [22], polyphenolic compounds [23]
CO2-philic compounds [24] and many more. However, the solubility data reported in the literatures
are often extremely divergent and inconclusive.

In this work, the dynamic method was used to experimentally determine the solubility of grape seed
oil in SC-CO2 in the range of pressure and temperature of practical importance (between 20 and 50
MPa and between 313 and 343 K). It is worth highlighting that only a limited number of literature
reports are available which deal with solubility of grape seed oil; in addition, they experimentally
investigated the solubility in restricted ranges of pressure and temperature.

The bulk majority of the scientific papers determine the solubility of solutes in SC-CO2 through
theoretical models. Therefore, testing the predictive power of solubility models commonly used in
the literature taking as a reference a common and extended basis of experimental data is of extreme
importance. In this regard, the experimental data were modeled using eight density-based models
and a thermodynamic model based on Peng-Robinson Equation of State (EoS). The models are
compared and discussed in terms of their effectiveness in predicting the experimental data.

2. Experimental
2.1 Materials, equipment and extraction procedure

CO2 (4.0 type, purity greater than 99.99%) purchased from Messer (Padua, Italy) was used as
supercritical solvent. Dry grape seeds (humidity 1-2 %) of Moscato cultivar were milled by a
grinder (Sunbeam Osterizer blender, Boca Raton, USA) and extracted using the supercritical
equipment (PRORAS, Rome, Italy) as detailed in [4] at a pressure of 50 MPa and a temperature of
323 K. A minor modification was made by adding a mini Cori-Flow digital mass flow meter
(Bronkhorst, Ruurlo, The Netherlands) placed on the liquid CO2 line, upstream of the CO2 pump, as
discussed in [2]. The extracted grape seed oil was stored at ambient temperature in tightly closed
dark glass vials sealed with Parafilm before being used in the solubility determination tests. Lipids
composition of Moscato grape seed oil as established by 1 H-NMR quantitative analysis resulted the
following (data in mole fraction ± SD): triacylglycerols: 0.982 ± 0.005; diacylglycerols: 0.0110 ±
0.0003; sterols: 0.0050 ± 0.0002; oxidized lipids: 0.0020 ± 0.0003 [2]. Fatty acid composition of
Moscato grape seed oil is reported in table 1 [2]. The major triacylglycerols found were: PLL
(16:0,18:2,18:2), POL (16:0,18:1,18:2), POO (16:0,18:1,18:1), LLL (18:2,18:2,18:2), OLL
(18:1,18:2,18:2), OOL (18:1,18:1,18:2), and finally OOO (18:1,18:1,18:1) [2].

Table 1

2.2 Solubility determination procedure

The solubility of grape seed oil at different temperatures and pressures was determined by
thoroughly mixing 5 g of oil with 145 g of glass beads (diameter: 1.05 mm) in an extractor of 0.1 L
volume (extractor assembly reported in [25]) and re-extracting the oil by SC-CO2. The amount of
oil used in the experiments (5 g) was fixed to allow a significant wetting of the glass beads, but
avoiding that the oil came down along the cylindrical extractor by gravity, which would
compromise the reliability of the solubility data, as evidenced by Sovovà et al. [8] (see Section 4.1.
for details).

The pressure and temperature were controlled during the tests with an accuracy of ±0.1 MPa and ±1
K, respectively. The cumulative CO2 consumption during the tests was measured and recorded by
the digital mass flow meter. The extracted oil dropped down in a beaker placed over a precision
balance (Mettler Toledo PL-203-S, Milan, Italy) and its amount was recorded during the tests. The
range of flow rates required to saturate the solvent was established by conducting several tests at the
same pressure and temperature and varying the solvent flow rate (see Section 4.1).

The solubility data for a given pressure and temperature was calculated as the initial slope of the
plot where the amount of extracted oil (y-axis) was plotted against the CO2 consumed (x-axis).

As a general rule, the static method is preferable with respect to the dynamic method for measuring
the solubility of a liquid consisting of a mixture of components. Actually, in the dynamic method
differences in composition of the solute saturating the solvent will occur during the course of the
dynamic test. Made this premise, in the present case the use of the dynamic method seems adequate:
the oil consists of 98.2 % triacylglycerols (and 1.1 % diacylglycerols), so that the presence of the
other minor constituents should not affect to an appreciable extent the value measured at the
beginning of the dynamic test and assumed as the solubility value of grape seed oil in SC-CO2.

3. Modeling

The modeling of solubility in SC-CO2 generally follows two approaches: one option is resorting to
density-based correlations; another is using thermodynamic models based on EoS. The models were
fit to the experimental data using MATLAB R2014a by nonlinear optimization function lsqcurvefit.
The deviation between model prediction and experimental data was quantified using root mean
square error (RMSE) and percent average absolute relative deviation (AARD).

3.1 Density-based models

In the scientific literature, there are at least three broad categories of density-based models.

The first group includes those models that are based on the law of mass action applied to a solvation
reaction. Notable examples in this group are the Chrastil’s model [26] and its modifications which
foresee a linear relationship between the logarithm of solubility and the logarithm of solvent
density. Some of the important modifications comprise the models by Adachi and Lu [27], del Valle
and Aguilera [28], and Sparks et al. [29].

The second category consists of those models that are based on the theory of dilute solutions. This
group of models assumes an iso-fugacity condition between the solute in the condensed phase and
the solute solubilized in the supercritical phase. Such group includes the models by Kumar and
Johnston [30], Bartle et al. [31], and Mendez-Santiago and Teja [32].
The third group contains models which are purely empirical in nature and which correlate the
solubility with pressure and temperature simply by using polynomial functions. An important
example of this last class is the Yu et al.’s model [33] which has several modifications in the
literature.

The model developed by Chrastil [26] is derived based on the association reaction between solute
and solvent. It is assumed that, at equilibrium, one molecule of solute A associates with k molecules
of solvent B, to form a solute-solvent complex. The final expression of the solubility
(kg·m-3) in
the supercritical phase is given by equation (1):

ln
⁄kg · m
  = ∆⁄ +  ln ⁄kg · m
 + ln +   ⁄kg · kmol
  −
 ln ⁄kg · kmol
  + q (1)

where ∆ is the total heat of reaction (J·mol-1), i.e. the sum of heat of solvation and heat of
vaporization (∆ = ∆ !" +∆"#$ ,  is the universal gas constant (J·mol-1·K-1),  is the system
temperature (K),  is the solvent density (kg·m-3),  and  are the molecular weight of,
respectively, solute and solvent (kg·kmol-1),  is the association number, and q is a constant.
Equation (1) can be rewritten as:

ln
⁄kg · m
  =  ln⁄kg · m
 + % ⁄ + & (2)

where % = ∆ ⁄ and & = ln +   ⁄kg · kmol

 −  ln ⁄kg · kmol
  + q

Since it was first introduced, Chrastil’s model [26] underwent several empirical modifications
including those by Adachi and Lu [27], del Valle and Aguilera [28], and Sparks and co-workers
[29].

Adachi and Lu [27] applied Chrastil’s model to over 37 different solutes in supercritical CO2 and
ethylene. They argued that, in the application of Chrastil’s model, the density of the solvent plays a
vital role therefore the association number  must be a function of density. They proposed an
empirical modification that correlates  and  as reported in equation (3):

 = '( + ' +  ) ') (3)

where '( , ' and ') are constant parameters which are determined by fitting the model to
experimental data. This correlation adds two more additional adjustable parameters to the Chrastil’s
model.
According to del Valle and Aguilera [28], the major drawback of Chrastil’s model is its
applicability over a restricted temperature range. Hence, an empirical modification was introduced
to compensate for the variation of ∆"#$ with temperature. Consequently, Chrastil’s model was
modified as reported in equation (4) by adding one more temperature dependent parameter to the
original Chrastil’s model, i.e. introducing the parameter *.

ln
⁄kg · m
  =  ln⁄kg · m
 + % ⁄ + * ⁄ ) + & (4)

Sparks et al. [29] applied six density-based models to six solute-supercritical systems. They
ascertained that the modifications by Adachi-Lu and del Valle–Aguilera improved indeed the
performance of Chrastil’s model. However, cases exist where one modification is better than the
other is, and opposite cases. Therefore, Sparks et al. [29] proposed a further modification by
incorporating both the Adachi-Lu and del Valle–Aguilera modifications in a single equation. They
recommend that the third term in the association constant correlation by Adachi and Lu (equation 3)
can be ignored without compromising the performance of the solubility model. Therefore, Sparks et
al.’s model [29] can be used with five or six adjustable parameters.

The Kumar and Johnston’s model [30] was developed under the assumption that the solute is
incompressible and, at equilibrium, the chemical potentials of the solute in the condensed phase and
in the supercritical phase are equal. Consequently, equation (5) was developed to correlate
solubility and density.

123 45 1 υs2 =
>2
ln +2 = −C + ./ 0 :+ −0 : ln D (5)
67 89 89 ?@89 6
ABC

In equation (5), +) is the mole fraction of solute in supercritical phase, C is a constant, E) is the
vapor pressure of solute (MPa), F is the critical density of solvent (kg·m-3), E is pressure (MPa), G)
is the molar volume of solute (m3·mol-1), GH) is the partial molar volume of solute in supercritical
phase (m3·mol-1), I 9 is isothermal compressibility (MPa-1), D is the reduced density of solvent
(D = ⁄J ). Equation (5) can be rewritten as:

L
ln +2 = K + 9 + M ln D (6)

123 45 1Gs2 =
>2
Where K = ./ 0 : − M , N = and M = − 0 :
67 89 8 ?@ 89 6
ABC

Bartle et al [31] proposed a Kumar and Johnston [30] type model, introducing the concept of
enhancement factor, defined as the ratio of actual solubility and ideal solubility. Reference pressure
and density were also introduced in their work as shown in equation (7):
 E L
ln Oy2 T = K + + MU − DVW X (7)
EQRS 9

where EDVW is the reference pressure (taken as 0.1 MPa) and DVW is the reference density (taken as
700 kg·m-3). Importantly, K, N and M in the Bartle et al.’s model are not the same as those in the
Kumar and Johnston’s model.

Mendez-Santiago and Teja [32] developed another Kumar and Johnston [30] type model, using the
concept of infinite dilution. The model is semi-empirical in nature and foresees a linear relationship
between the logarithm of enhancement factor and density. The model has three adjustable
parameters as shown in equation (8).

lnU+2E⁄MPaX = K + N + M (8)

Yu el al. [33] argued that the solubility of solutes in SC-CO2 follows a curvilinear behavior with
pressure at a constant temperature, with temperature at a constant pressure, and with the interaction
of these two thermodynamic variables. Thus, a second order polynomial correlation with both
temperature and pressure was proposed, as shown in equation (9):

+) = 1 − K + NE + ME) + ] + ^ ) + _E+)  (9)

where K, N, M, ], ^ and _ are constants.

Yu et al.’s model was later modified by Gordillo et al. [34] and Jouyban et al. [35]. The final form
of such modified versions is the same as that of the original Yu et al.’s model [33], the difference
being the reduction in one or two parameters. Because of that, Gordillo et al.’ model [34] and
Jouyban et al.’s model [35] are not discussed further in this work.

3.2 Thermodynamic model

A widely accepted thermodynamic approach for the determination of solubility in supercritical

fluids consists in the use of Peng-Robinson EoS with classical van der Waals’ mixing rules
[14,15,17,21,36,37]. Such thermodynamic approach is based on the iso-fugacity condition between
the solute in the condensed phase and the solute in the supercritical phase. Common simplifying
assumptions consider the solute pure, incompressible, and having low vapor pressure [15,21]. For
binary systems, the final expression of solubility in supercritical phase is given by equation (10)
[38]:

1` i`21
12` 
y) = 1a2bcd exp h jk
l (10)
2
where mnop
) is the fugacity coefficient of solute in supercritical phase and the other parameters are
as defined in Section 3.1. The fugacity coefficient is determined from Peng-Robinson EoS [39] with
van der Waals’ mixing rules and the binary interaction coefficients (qr and .qr ) are used as
adjustable parameters to fit the model with experimental data. The physical properties used in this
work were taken from Yu et al. [33], by approximating the oil as consisting of triolein. The vapor
pressure at the different temperatures was estimated by the Wagner’s vapor pressure equation [40].

4. Results And Discussion

4.1 Solubility data

For a correct determination of solubility when using the dynamic method, one of the main
challenges is to make sure that the solvent leaving the extractor is saturated by the solute. The
technique frequently used in the literature to guarantee solvent saturation foresees to conduct
repeated experiments under the same operating conditions (, E) by varying the solvent flow rate.
Under saturation conditions, the initial slope of the extraction kinetics curves, i.e. the plot of mass
of solute extracted versus mass of solvent consumed, must overlap. Figure 1 shows the kinetics of
extraction of grape seed oil from the surface of glass beads at a pressure of 50 MPa and temperature
of 323 K. It is evident from figure 1 as the extraction curves overlap in the tested flow rate range of
0.11-0.15 g s-1 indicating that the solvent is indeed saturated. A similar behavior was observed in
other process conditions: operating the apparatus at a solvent flow rate of about 0.10 g s-1 allowed to
obtain consistent solubility data in the whole range of pressure and temperature investigated.

Figure 1

Interestingly, when extracting ground seed particles, the solvent flow rate allowing for solvent
saturation resulted lower than the value reported for glass beads. For instance, Duba and Fiori [25]
obtained an initial slope of the extraction kinetics curve resembling the solubility value (table 2,  =
313 K and E = 35 MPa) when extracting grape seed particles at a low flow rate of 0.08 g s-1, while
at higher flow rates (0.12-0.17 g s-1) the solvent was not saturated.

Most of the works in the literature dealing with the determination of solubility of vegetable oils in
SC-CO2 uses directly ground seed matrix instead of glass beads to observe solvent saturation. The
main advantage of using glass beads over the ground vegetable substrate is the elimination of
internal mass transfer resistances while preserving the packed bed structure typical of extractors.
However, care must be taken and glass beads should be used with caution when determining the
solubility. Sovovà et al. [8] found out that the maximum amount of oil to be added on the surface of
glass beads was 0.5 gram when using an extractor volume of 8 cm3. When the amount of oil fed to
the extractor was greater than the recommended value, it was observed that part of the oil flowed
down the extractor by gravity and this resulted in misleading solubility values [8]. The extractor
used in this work was more than ten times (100 cm3) the volume of the extractor employed by
Sovovà et al. [8]. Consequently, the mass of oil used here was ten times (5 g) the proposed amount.
Before starting the experimental campaign of this work, a practical check was performed as an
additional verification to avert the problem of the possible down-flow of the oil in the extractor. 5 g
of oil were thoroughly mixed with glass bead in the 0.1 L extractor basket, which was then let to
stand for an hour on a white paper: this confirmed experimentally that natural oil down flow did not
occur.

Table 2 shows the experimental values of solubility of grape seed oil over the range of pressure and
temperature investigated. Each data point represents the average of at least three consecutive points
on the linear part of the extraction kinetics curve. On the one hand, with an increase in pressure at
constant temperature, the solubility increases in line with the increase in solvent density, as
expected. On the other hand, the effect of temperature is different for low and high pressure due to
two competing factors, i.e. the solute vapor pressure and the solvent density [41]. At low pressure,
increasing the temperature has a negative effect on the solubility, indicating the density effect
dominates the vapor pressure effect. Conversely, at high pressure an increase in temperature
enhances solubility, indicating the solute vapor pressure effect prevails.

Table 2

Table 2 shows also the range of vegetable oil solubility computed according the so-called ‘general
model’ proposed by del Valle et al. [16]. The general structure of this model is the same as that of
Sparks et al.’s model [29], except that the model parameters are optimized for a wider range of
vegetable oils. del Valle et al. [16] claimed that the model is capable of predicting vegetable oil
solubility within the range of ±40% for combinations of temperature and pressure of practical
importance. Table 2 testifies that the reported experimental solubility data are within the range of
predicted values of the general model except for 20 MPa at temperature of 333 and 343 K.
According to del Valle et al. [16], the proposed model does not apply for low solubility (≤1 kg·m-3)
unless the system pressure is greater than 21 MPa: this is very likely the reason for the mismatching
of the data at 20 MPa at 333 and 343 K.

In a previous work by Fiori [4], grape seed oil solubility values were reported for  = 313 K. Such
values were higher than the values reported in this work and were close to the upper limit of the
±40% range as stated by del Valle et al. [16]. The authors believe that the main cause of this was the
way the CO2 flow was quantified. In the original equipment configuration, the CO2 flow rate was
measured after expansion and separation of the oil (reference can be done to the P&Id of the
extraction equipment reported in [4]). It cannot be excluded that some CO2 left the expansion vessel
through the oil recovery line, which made the measured CO2 flow rate slightly lower than the actual
value and hence the solubility apparently higher. To overcome this problem, a plant modification
was introduced as discussed in Section 2.1: in the current work, the CO2 flow rate was measured in
the pipeline upstream of the liquid CO2 pump rather than in the CO2 vent line.

Temperature and pressure are the main parameters affecting solubility. Columns 5-8 in table 2 show
the standard uncertainty components as a result of repeatability t, temperature, pressure, and
combined expanded uncertainty uJ , determined according to NIST Thermodynamic Research
Center (TRC) recommendations as discussed by Chirico et al. [42] and Dong et al. [43]. The
uncertainty as a result of CO2 flow rate was neglected as this has little effect in the range of flow
rates of this experimental work: Figure 1. The sensitivity coefficients, i.e. the partial derivative of
solubility with respect to pressure at constant temperature, and the partial derivative of solubility
with respect to temperature at constant pressure, were determined from the plot of experimental
solubility versus pressure and temperature curves in the appropriate range. The sensitivity values
were then multiplied with the pressure and temperature standard variation of ±0.1 MPa and ±1 K to
get standard uncertainty of the components in solubility unit: tE and t, respectively. The
individual uncertainty components were then combined according to the ISO Guide to the
Expression of Uncertainty in Measurement (GUM) to get a combined expanded uncertainty uJ with
95% confidence level.

4.2 Correlation of Solubility

The solubility correlations by Chrastil’s model and its modifications are shown in figure 2. The
model adjustable parameters along with the deviations from experimental data are presented in table
3. As discussed in Section 3.1, the del Valle and Aguilera’s, Adachi and Lu’s and Sparks et al.’s
modifications have, respectively, one, two and three additional adjustable parameters in respect to
the original Chrastil’s model. As can be seen in table 3, the model adjustable parameters, , % and &
of the Chrastil’s model and the del Valle and Aguilera’s modification are extremely similar.
Surprisingly, the additional parameter introduced to offset the effect of temperature in the del Valle
and Aguilera’s modification takes a constant value of 0.20 for all the temperatures. Note also that,
because of the small value of *, the term *⁄ ) becomes insignificant relatively to the magnitude
of ./
, which makes the del Valle and Aguilera’s and Chrastil’s models to respond in the same way
to the experimental data. Therefore, under the range of temperatures and pressures of this study, the
del Valle and Aguilera’s modification provides only limited improvements to the original Chrastil’s
model.

Figure 2

Table 3

The Adachi and Lu’s modification was proposed to introduce the dependence of the association
constant  on solvent density. It can be observed from the trend of adjustable parameters in table 3
that the coefficient of square density in the proposed correlation is zero for all the temperatures.
Sparks et al. [29] hinted that a linear relationship between  and  suffices, and therefore the
coefficient ') can be set equal to zero without compromising the performance of the model.
Accordingly, the number of adjustable parameters can be reduced. Furthermore, the Adachi and
Lu’s modification made the coefficient of temperature % in the Chrastil’s model constant. This
argument is also true for the Sparks et al.’s modification, which responds to the experimental data in
exactly the same way as the Adachi and Lu’s modification. The Sparks et al.’s modification
includes both the modifications by del Valle and Aguilera [28] and Adachi and Lu [27], as
discussed in Section 3.1. This makes the model to behave in the same way as the combination of
both predecessors. The parameter * is constant like in the del Valle and Aguilera’ model (but it has
a different value), while % is constant as in the Adachi and Lu’s model. Unlike del Valle and
Aguilera’s modification, the use of Adachi and Lu’s and Sparks et al.’s modifications results in
completely different values of adjustable parameters % and & in respect to the original Chrastil’s
model. Therefore, it is impossible to infer the parameters of the Chrastil’s model from the Adachi
and Lu’s (and Sparks et al.’s) model parameters. Consequently, it is difficult to assign a physical
meaning to the Adachi and Lu’s and Sparks et al.’s model parameters, indicating the modifications
result in different empirical models. Both the Adachi and Lu’s and Sparks et al.’s modifications
reduced the RMSE and percent AARD of both the original Chrastil’s model and the del Valle and
Aguilera’s model (table 3), at expense of an increased number of adjustable parameters.

The solubility correlations of the second class of density-based models are shown in figure 3. This
category includes the Kumar and Johnston’s [30], Bartle et al.’s [31] and Mendez-Santiago and
Teja’s [32] models as discussed in Section 3.1. All three models have the same number of
adjustable parameters (three), but they differ from each other by the way in which the relationship
between solubility and density is defined. For instance, the difference between Mendez-Santiago
and Teja’s and Bartle et al.’s models is in the way the temperature dependence of the solubility is
expressed. The Bartle et al.’s model introduced the concept of reference pressure (0.1 MPa) and
reference density (700 kg·m-3) to reduce the error which may be incurred because of the variation of
parameters with density. Both of these models have the same form as Kumar and Johnston’s model:
see equations (6-8). However, it must be noted that the physical significance of the coefficients is
different in the three models. As shown in Table 4, Bartle et al.’s and Mendez-Santiago and Teja’s
models react to the experimental data in exactly similar fashion, with the same values of RMSE and
AARD. By comparison, Kumar and Johnston’s model presents relatively lower values of AARD.
Importantly, looking at these three density-based models, only Kumar and Johnston’s model
presents a direct relationship between the logarithm of solubility and the logarithm of solvent
density.

Figure 3

Table 4

The Yu et al.’s empirical model considers a curvilinear relationship between solubility and solvent
physical properties (temperature and pressure), with second order polynomial. The model fits the
experimental data relatively better than the other density-based models, as shown in table 4. The
major drawback of Yu et al.’s model is the large number of adjustable parameters besides the lack
of any physical significance of the fitting parameters. As can be seen in table 4, the coefficient of
E) is zero. A zero value for an adjustable parameter was also observed by Yu el al. [33] when
dealing with the solubility of rapeseed oil in SC-CO2, but in their case not for the coefficient of E)
but rather for that of E. Obtaining a zero value for certain circumstances does not necessarily
mean the number of model adjustable parameters can be reduced.

The prediction of solubility with Peng-Robinson EoS is presented in figure 4. Interestingly, the
predictive power of Peng-Robinson EoS evaluated in terms of RMSE and AARD (table 5) is
comparable, if even a little worse, to that of density-based models. In this case, the fitting
parameters are the binary interaction coefficients vw and .vw , as reported in details by
Mukhopadhyay [38]. The values of both adjustable parameters are relatively constant with
temperature (0.243±0.016 for vw and 0.138±0.006 for .vw ), which makes it possible to determine the
solubility only from physical properties of solute and solvent. However, the fact that the physical
properties data are only approximately known for most of the solutes of practical importance is the
major drawback of this approach.

Figure 4

Table 5
5. Conclusions

The solubility of grape seed oil was determined in a wide range of temperature and pressure,
namely for the temperatures  = (313, 323, 333, 343) K and for the pressures E = (20, 35, 40, 50)
MPa. The results show that the solubility increases with the increase in pressure at constant
temperature, while the effect of temperature is different for low and high pressure. At low pressure
(20 MPa) the solubility decreases with the increase in temperature. Conversely, at high pressure (≥
35 MPa) it increases with the increase in temperature, as a result of the relative importance of the
solute vapor pressure and the solvent density. The experimental data were modeled by eight
density-based models, which are widely used in the literatures: namely the models by Chrastil, del
Valle and Aguilera, Adachi and Lu, Sparks et al., Kumar and Johnston, Bartle et al., Mendez-
Santiago and Teja, and Yu et al., as well as with a thermodynamic model based Peng–Robinson
EoS. The result shows that all the models are capable of predicting the solubility of grape seed oil in
SC-CO2 to a reasonable degree. However, best fitting alone does not provide a guarantee with
respect to the actual effectiveness of a model. A good model should sufficiently describe the
underlining physical solubilization phenomena, should be reasonably simple and should contain a
few adjustable parameters. To this regard, the models by Chrastil, del Valle and Aguilera, Kumar
and Johnston and that based on Peng–Robinson EoS represent effective and physically-based tools
for predicting the solubility of oil in SC-CO2.

Acknowledgment

Research supported by AGER (project number 2010-2222).

References

[1] L. Fernandes, S. Casal, R. Cruz, J.A. Pereira, E. Ramalhosa, Seed oils of ten traditional
Portuguese grape varieties with interesting chemical and antioxidant properties, Food Res.
Int. 50 (2013) 161–166. doi:10.1016/j.foodres.2012.09.039

[2] L. Fiori, V. Lavelli, K.S. Duba, P.S.C. Sri Harsha, H. Ben Mohamed, G. Guella, Supercritical
CO2 extraction of oil from seeds of six grape cultivars: Modeling of mass transfer kinetics
and evaluation of lipid profiles and tocol contents, J. Supercrit. Fluids. 94 (2014) 71–80.
doi:10.1016/j.supflu.2014.06.021

[3] A. Hanganu, M. C. Todaşcă, N.A. Chira, M. Maganu, S. Roşca, The compositional

characterisation of Romanian grape seed oils using spectroscopic methods, Food Chem. 134
(2012) 2453–2458. doi:10.1016/j.foodchem.2012.04.048

[4] L. Fiori, Grape seed oil supercritical extraction kinetic and solubility data: Critical approach
and modeling, J. Supercrit. Fluids. 43 (2007) 43–54. doi:10.1016/j.supflu.2007.04.009
[5] C.P. Passos, R.M. Silva, F.A. Da Silva, M.A. Coimbra, C.M. Silva, Enhancement of the
supercritical fluid extraction of grape seed oil by using enzymatically pre-treated seed, J.
Supercrit. Fluids. 48 (2009) 225–229. doi:10.1016/j.supflu.2008.11.001

[6] Ž. Knez, E. Markočič, M. Leitgeb, M. Primožič, M. Knez Hrnčič, M. Škerget, Industrial

applications of supercritical fluids: A review, Energy. (2014) 1–9.
doi:10.1016/j.energy.2014.07.044

[7] I. Dalmolin, A. A. Rigo, M.L. Corazza, P.M. Ndiaye, M.A. A. Meireles, E. A. C. Batista,
Phase behaviour and thermodynamic modelling for the system (grape seed oil+carbon
dioxide+ethanol) at high pressures, J. Chem. Thermodyn. 68 (2014) 71–74.
doi:10.1016/j.jct.2013.09.002

[8] H. Sovovà, M. Zarevu, M. Vacek, K. Stra, Solubility of two vegetable oils in supercritical
CO2, J. Supercrit. Fluids. 20 (2001) 15–28. doi:10.1016/S0896-8446(01)00057-2

[9] M. Hojjati, Y. Yamini, M. Khajeh, A. Vatanara, Solubility of some statin drugs in

supercritical carbon dioxide and representing the solute solubility data with several density-
based correlations, J. Supercrit. Fluids. 41 (2007) 187–194. doi:10.1016/j.supflu.2006.10.006

[10] H. Li, D. Jia, S. Li, R. Liu, Correlating and predicting the solubilities of structurally similar
organic solid compounds in supercritical CO2 using the compressed gas model and the
reference solubilites, Fluid Phase Equilib. 350 (2013) 13–26. doi:10.1016/j.fluid.2013.04.010

[11] Y. Yamini, M. Moradi, Measurement and correlation of antifungal drugs solubility in pure
supercritical CO2 using semiempirical models, J. Chem. Thermodyn. 43 (2011) 1091–1096.
doi:10.1016/j.jct.2011.02.020

[12] A. Galia, A. Argentino, O. Scialdone, G. Filardo, A new simple static method for the
determination of solubilities of condensed compounds in supercritical fluids, J. Supercrit.
Fluids. 24 (2002) 7–17. doi:10.1016/S0896-8446(02)00009-8

[13] M. Hosseini Anvari, G. Pazuki, A study on the predictive capability of the SAFT-VR
equation of state for solubility of solids in supercritical CO2, J. Supercrit. Fluids. 90 (2014)
73–83. doi:10.1016/j.supflu.2014.03.005

[14] M. Khamda, M.H. Hosseini, M. Rezaee, Measurement and correlation solubility of cefixime
trihydrate and oxymetholone in supercritical carbon dioxide (CO2), J. Supercrit. Fluids. 73
(2013) 130–137. doi:10.1016/j.supflu.2012.09.006

[15] S. Zhao, D. Zhang, An experimental investigation into the solubility of Moringa oleifera oil
in supercritical carbon dioxide, J. Food Eng. 138 (2014) 1–10.
doi:10.1016/j.jfoodeng.2014.03.031

[16] J.M. del Valle, J.C. de la Fuente, E. Uquiche, A refined equation for predicting the solubility
of vegetable oils in high-pressure CO2, J. Supercrit. Fluids. 67 (2012) 60–70.
doi:10.1016/j.supflu.2012.02.004

[17] C. Garlapati, G. Madras, Solubilities of palmitic and stearic fatty acids in supercritical carbon
dioxide, J. Chem. Thermodyn. 42 (2010) 193–197. doi:10.1016/j.jct.2009.08.001
[18] S.N. Reddy, G. Madras, An association and Wilson activity coefficient model for solubilities
of aromatic solid pollutants in supercritical carbon dioxide, Thermochim. Acta. 541 (2012)
49–56. doi:10.1016/j.tca.2012.04.025

[19] V.F. Cabral, W.L.F. Santos, E.C. Muniz, A. F. Rubira, L. Cardozo-Filho, Correlation of dye
solubility in supercritical carbon dioxide, J. Supercrit. Fluids. 40 (2007) 163–169.
doi:10.1016/j.supflu.2006.05.004

[20] P. Coimbra, M.H. Gil, C.M.M. Duarte, B.M. Heron, H.C. de Sousa, Solubility of a
spiroindolinonaphthoxazine photochromic dye in supercritical carbon dioxide: Experimental
determination and correlation, Fluid Phase Equilib. 238 (2005) 120–128.
doi:10.1016/j.fluid.2005.09.024

[21] R.S. Alwi, T. Tanaka, K. Tamura, Measurement and correlation of solubility of

anthraquinone dyestuffs in supercritical carbon dioxide, J. Chem. Thermodyn. 74 (2014)
119–125. doi:10.1016/j.jct.2014.01.015

[22] K. A. Araus, J.M. del Valle, P.S. Robert, J.C. de la Fuente, Effect of triolein addition on the
solubility of capsanthin in supercritical carbon dioxide, J. Chem. Thermodyn. 51 (2012) 190–
194. doi:10.1016/j.jct.2012.02.030

[23] A.L. Cabrera, A.R. Toledo, J.M. del Valle, J.C. de la Fuente, Measuring and validation for
isothermal solubility data of solid 2-(3,4-Dimethoxyphenyl)-5,6,7,8-tetramethoxychromen-4-
one (nobiletin) in supercritical carbon dioxide, J. Chem. Thermodyn. 91 (2015) 378–383.
doi:10.1016/j.jct.2015.08.018

[24] J. Peng, H.-J. Yang, H. Yang, Solubilities and partial molar volumes of three bis(2-
ethoxyethyl) ethanedioate derivatives in supercritical carbon dioxide, J. Chem. Thermodyn.
83 (2015) 33–42. doi:10.1016/j.jct.2014.11.009

[25] K.S. Duba, L. Fiori, Supercritical CO2 extraction of grape seed oil: Effect of process
parameters on the extraction kinetics, J. Supercrit. Fluids. 98 (2015) 33–43.
doi:10.1016/j.supflu.2014.12.021

[26] J. Chrastll, Solubility of solids and liquids in supercritical gases, J. Phys. Chem. 86 (1982)
3016–3021. DOI: 10.1021/j100212a041

[27] Y. Adachi, B.C.Y. Lu, Supercritical fluid extraction with carbon dioxide and ethylen, Fluid
Phase Equilib. 14 (1983) 147–156. doi:10.1016/0378-3812(83)80120-4

[28] J.M. del Valle, J.M. Aguilera, An improved equation for predicting the solubility of
vegetable oils in supercritical carbon dioxide, Ind. Eng. Chem. Res. 27 (1988) 1551–1553.
DOI: 10.1021/ie00080a036

[29] D.L. Sparks, R. Hernandez, L.A. Estévez, Evaluation of density-based models for the
solubility of solids in supercritical carbon dioxide and formulation of a new model, Chem.
Eng. Sci. 63 (2008) 4292–4301. doi:10.1016/j.ces.2008.05.031.

[30] S.K. Kumar, P. Johnston, Modelling the solubility of solids in supercritical fluids with
density as the independent variable, J. Supercrit. Fluids. 1 (1988) 15–22. doi:10.1016/0896-
8446(88)90005-8
[31] K.D. Bartle, A.A. Clofford, S.A. Jafar, G.F. Shilstone, Solubilities of solids and liquids of
low volatility in supercritical carbon dioxide, J. Phys. Chem. Ref. Data. 20 (1990) 1–45.
doi:10.1063/1.555893

[32] J. Mendez-Santiago, A.S. Teja, The solubility of solids in supercritical fluids, Fluid Phase
Equilib. 158-160 (1999) 501–510. doi:10.1016/S0378-3812(99)00154-5

[33] Z. Yu, S.S.H. Rizvi, J.A. Zollweg, Solubilities of fatty acids, fatty acid esters, triglycerides,
and fats and oils in supercritical carbon dioxide, J. Supercrit. Fluids. 7 (1994) 51–59.
doi:10.1016/0896-8446(94)90006-X

[34] M.D. Gordillo, M.A. Blanco, A. Molero, E.M. De la Ossa, Solubility of the antibiotic
Penicillin G in supercritical carbon dioxide, J. Supercrit. Fluids. 15 (1999) 183–190.
doi:10.1016/S0896-8446(99)00008-X

[35] A. Jouyban, H. Chan, N.R. Foster, Mathematical representation of solute solubility in

supercritical carbon dioxide using empirical expressions, J. Supercrit. Fluids. 24 (2006) 19–
35. doi:10.1016/S0896-8446(02)00015-3

[36] C.I. Park, M.S. Shin, H. Kim, Solubility of climbazole and triclocarban in supercritical
carbon dioxide: Measurement and correlation, J. Chem. Thermodyn. 41 (2009) 30–34.
doi:10.1016/j.jct.2008.08.009.

[37] L. Fiori, M. Manfrini, D. Castello, Supercritical CO2 fractionation of omega-3 lipids from
fish by-products: Plant and process design, modeling, economic feasibility, Food Bioprod.
Process. 92 (2014) 120–132. doi:10.1016/j.fbp.2014.01.001

[38] M. Mukhopadhyay, Natural Extracts Using Supercritical Carbon Dioxide, 1st ed., CRC Press
LLC, Filorida, 2000.

[39] D. Peng, D.B. Robinson, A new two-constant equation of state, Ind. Eng. Chem.
Fundamenals. 15 (1976) 59–64. DOI: 10.1021/i160057a011

[40] D.W. Green, R.H. Perry, Perry’s Chemical Engineering Handbook, 8th ed., McGraw-Hill,
New York, 2007.

[41] Z. Huang, S. Kawi, Y.. Chiew, Solubility of cholesterol and its esters in supercritical carbon
dioxide with and without cosolvents, J. Supercrit. Fluids. 30 (2004) 25–39.
doi:10.1016/S0896-8446(03)00116-5.

[42] R.D. Chirico, M. Frenkel, V. V. Diky, K.N. Marsh, R.C. Wilhoit, ThermoML-An XML-
based approach for storage and exchange of experimental and critically evaluated
thermophysical and thermochemical property data. 2. Uncertainties, J. Chem. Eng. Data 48
(2003) 1344–1359. DOI: 10.1021/je034088i

[43] Q. Dong, R.D. Chirico, X. Yan, X. Hong, M. Frenkel, Uncertainty reporting for experimental
thermodynamic properties, J. Chem. Eng. Data 50 (2005) 546–550. DOI: 10.1021/je049673d

[44] NIST Chemistry WebBook. http://webbook.nist.gov/chemistry/ (accessed 01.04.2016).

Figure captions

Figure 1. Extraction kinetics curves at varying CO2 flow rate: mass of oil extracted (xyz{ ) versus
mass of CO2 utilized (xoy) ) (E = 50 MPa;  = 323 K).

Figure 2. Solubility (
) versus CO2 density (): experimental data and modeling curves for: (a)
Chrastil’s model [26]; (b) del Valle and Aguilera’s model [28] ; (c) Adachi and Lu’s model [27]; (d)
Sparks et al.’s model [29].

Figure 3. Solubility, expressed in terms of mole fraction of solute in supercritical phase (+) ); E is
pressure; EDVW is reference pressure (0.1 MPa); ρ is CO2 density; DVW is reference density (700
kg·m-3): experimental data and modeling curves. (a) Kumar and Johnston’s model [30]; (b) Bartle et
al.’s model [31]; (c) Mendez-Santiago and Teja’s model [32]; (d) Yu et al.’s model [33].

Figure 4. Solubility, expressed in terms of mole fraction of solute in supercritical phase (+) ), versus
pressure (E): experimental data and modeling curves for the thermodynamic model with Peng-
Robinson EoS [38]. Sub-charts refer to different temperatures: (a)  = 313 K; (b)  = 323 K; (c) 
= 333 K; (d)  = 343 K.
Table 1

Table 1: Fatty acid composition (% of total fatty acids) from FAME GC-FID-MS analysis of Moscato grape seed oil. Data are expressed as mean ±
SD. Data from Fiori et al. [2].

.
C14:0 C16:0 C17:0 C18:0 C18:1 (-9) C18:2 (-6) C18:3 (-3) C20:0 C20:1 (-9) C20:2 (-6)
0.051± 0.003 8.89± 0.21 0.049± 0.001 2.84± 0.02 15.3± 0.1 71.0± 0.3 0.46± 0.01 0.14± 0.01 0.11± 0.01 0.041± 0.010
Table 2

Table 2: Solubility ( ) of grape seed oil in supercritical CO2 at various temperatures (T) and
pressures (P). Value of CO2 density ( ) from NIST [44]. is repeatability; and are
pressure and temperature dependent standard uncertainty components, respectively; is combined
-1
expanded uncertainty and is a coverage factor. The solubility is here reported in g·kg to allow
for a direct comparison with the solubility value by the “general model” by del Valle et al. [16].

/(g·kg-1) General
/K /MPa /(kg·m-3) /(g·kg-1) /(g·kg-1) /(g·kg-1) /(g·kg-1) model
/(g·kg-1)
313 20 840.61 4.20 0.05 0.030 0.013 0.12 2.17-5.07
35 935.34 8.60 0.09 0.030 0.065 0.23 6.23-14.53
40 956.56 10.40 0.16 0.030 0.078 0.36 7.59-17.71
50 991.73 13.00 0.30 0.030 0.105 0.64 10.15-23.68

323 20 785.16 3.53 0.44 0.033 0.013 0.88 1.60-3.72

35 899.77 9.50 0.27 0.033 0.065 0.56 6.59-15.39
40 923.81 11.06 0.10 0.033 0.078 0.26 8.49-19.81
50 962.89 13.40 0.21 0.033 0.105 0.47 12.27-28.64

333 20 724.63 3.12 0.05 0.039 0.013 0.13 1.05-2.45

35 863.49 10.00 0.13 0.039 0.065 0.30 6.62-15.45
40 890.64 12.00 0.23 0.039 0.078 0.49 9.06-21.15
50 933.94 14.60 0.12 0.039 0.105 0.33 14.27-33.29

343 20 660.04 2.91 0.30 0.044 0.013 0.61 0.64-1.450

35 826.66 10.60 0.25 0.044 0.065 0.52 6.34-14.79
40 857.20 12.70 0.63 0.044 0.078 1.27 9.27-21.63
50 904.98 16.10 0.82 0.044 0.105 1.66 15.99-37.32
Standard uncertainties are = 0.1 MPa and = 1 K.
Table 3

Table 3: value of model parameters ( ) from best fitting with experimental data at
various temperature ( ): model by Chrastil [26], del Valle and Aguilera [28], Adachi and Lu [27],
and Sparks et al. [29]. RMSE and AARD are root mean square error and percent average absolute
relative deviation, respectively.

Model /K RMSE AARD

·10-2 /%
Chrastil 313 7.62 0.040 -50.02 5.19 0.51
323 7.68 0.044 -50.17 8.63 0.99
333 7.22 0.059 -46.72 8.97 1.03
343 6.49 0.078 -41.48 6.14 0.69

del Valle & 313 7.62 0.040 -50.02 0.20 5.19 0.51
Aguilera 323 7.68 0.044 -50.18 0.20 8.63 0.99
333 7.22 0.059 -46.72 0.20 8.86 0.96
343 6.49 0.079 -41.48 0.20 6.14 0.69

Adachi & 313 0.499 0.14 -1.385 0.0024 0.00 4.44 0.35
323 0.499 -0.05 -2.385 0.0049 0.00 3.60 0.41
Lu
333 0.499 -0.03 -2.252 0.0049 0.00 2.23 0.23
343 0.499 0.10 -1.719 0.0039 0.00 0.64 0.07

Sparks et 313 0.499 0.14 0.50 -1.379 0.0024 0.00 4.44 0.35
al. 323 0.499 -0.04 0.50 -2.385 0.0049 0.00 3.60 0.41
333 0.499 -0.03 0.50 -2.252 0.0049 0.00 2.23 0.23
343 0.499 0.10 0.50 -1.719 0.0039 0.00 0.64 0.07
Standard uncertainty is = 1 K.
Table 4

Table 4: value of model parameters (A, B, C, D, E, F) from best fitting with experimental data at
various temperature ( ): model by Kumar and Johnston [30], Bartle et al. [31], Mendez-Santiago
and Teja [32], and Yu et al. [33]. RMSE and AARD are root mean square error and percent average
absolute relative deviation, respectively.

Model /K RMSE AARD

·10-2 /%
Kumar & 313 -12.31 -7.27 6.61 5.20 0.16
Johnston 323 -12.08 0.162 6.68 8.60 0.31
333 -11.47 0.165 6.22 8.99 0.33
343 -10.90 7.26 5.69 6.61 0.24

Bartle et 313 -5.02 0.18 0.013 5.63 0.96

al. 323 -4.37 -9.81 0.013 8.64 1.68
333 -3.76 10.19 0.012 8.63 1.85
343 -3.14 -9.81 0.011 6.25 1.35

Mendez- 313 -9.81 -14.30 4.16 5.63 0.96

Santiago & 323 10.12 -13.40 4.14 8.64 1.68
Teja 333 0.16 -12.06 3.96 8.63 1.85
343 -7.53 -10.89 3.79 6.25 1.35

Yu 313 0.20 -0.91 0.00 0.0029 0.199 -0.0006 2.28 0.96

et al. 323 0.20 -0.88 0.00 0.0027 0.199 -0.0006 1.28 0.38
333 0.20 -0.86 0.00 0.0026 0.199 -0.0006 1.81 0.52
343 0.20 -0.87 0.00 0.0025 0.199 -0.0006 1.42 0.39
Standard uncertainty is = 1 K.
Table 5

Table 5: value of model parameters ( ) from best fitting with experimental data at various
temperature ( ): thermodynamic model with Peng-Robinson EoS [38]. RMSE and AARD are root
mean square error and percent average absolute relative deviation, respectively.

Model /K RMSE AARD

·10-2 /%
PR-EOS 313 0.2407 0.1389 7.84 2.07
323 0.2248 0.1312 7.84 2.85
333 0.2423 0.1356 7.14 1.88
343 0.2647 0.1464 7.75 2.29
Standard uncertainty is = 1 K.
Highlights

• solubility of grape seed oil in SC-CO2 for E: 20-50 MPa and : 313-343 K
• experimental procedure: dynamic method and oil dispersed on the surface of glass beads
• eight density-based models and a thermodynamic model to fit the experimental data
• all the models predict the solubility of grape seed oil in SC-CO2 to a reasonable degree
• models by Chrastil, del Valle and Aguilera, Kumar and Johnston, and the thermodynamic
model are preferable