You are on page 1of 11


of Supercritical Fluids 98 (2015) 33–43

Contents lists available at ScienceDirect

The Journal of Supercritical Fluids

journal homepage:

Supercritical CO2 extraction of grape seed oil: Effect of process

parameters on the extraction kinetics
Kurabachew Simon Duba, Luca Fiori ∗
Department of Civil, Environmental and Mechanical Engineering (DICAM), University of Trento, via Mesiano 77, 38123 Trento, TN, Italy

a r t i c l e i n f o a b s t r a c t

Article history: The effect of the main process variables affecting the supercritical CO2 extraction of oil from seeds (namely
Received 24 September 2014 grape seeds) was investigated, both experimentally and through modeling. The dependency of the extrac-
Received in revised form tion kinetics on the variables more tested in the literature (pressure, temperature, particle size and solvent
24 December 2014
flow rate) was confirmed, and original trends were obtained for the less investigated variables, such as the
Accepted 26 December 2014
bed porosity ε and the extractor diameter to length ratio (D/L). The extraction kinetics did not depend on
Available online 3 January 2015
ε for 0.23 ≤ ε ≤ 0.41, while a further decrease in ε lowered the extraction rate, likely due to the occurrence
of channeling. The effect of a variable D/L ratio was studied letting constant the ratio of substrate mass
Supercritical CO2 extraction
to CO2 mass flow rate: the lower was D/L, the lower the specific CO2 consumption. Through modeling,
Extraction kinetics model the values of internal and external mass transfer parameters were calculated and critically discussed on
Grape seeds the basis of well-known literature correlations.
Seed oil © 2015 Elsevier B.V. All rights reserved.
BIC model
Mass transfer

1. Introduction conditions employed. The variety of cultivars and the environmen-

tal factors during harvesting year also play a significant role [6]. A
According to International Organization of Vine and Wine sta- wide range of oil yield (3.95–16.6%) of grape seeds from different
tistical report on world vitiviniculture 2013, the global wine cultivars is reported in the literature [6,7,9–11].
production of 2012 was evaluated to be 252 million hectoliters Traditionally seed oils are extracted either by organic solvent
[1]. European Union (EU-27) is the world leader in wine produc- or mechanical techniques. Organic solvent extraction gives bet-
tion, with almost half of the global vine-growing area and about ter extraction yield, but the technique requires solvent recovery
60% of production by volume with France, Italy and Spain being through distillation which may degrade thermally labile com-
the leading producers. Italy, the country where this research was pounds; moreover, the presence of traces of residual solvent in
conducted, stands in second position with a total production of 40 the final product makes the process less attractive from health
million hectoliters in 2012 [2]. In wine processing, over 0.3 kg of and environmental point of views. In mechanical extraction, even
solid by-products are produced per kg of fruit crushed [3]. The main though the product quality is superior (after proper filtration), the
by-product is grape marc which accounts for around two third of technique provides relatively lower yield. The use of the supercrit-
the solids (the rest being wine lees). Grape marc roughly consists of ical CO2 (SC-CO2 ) extraction process is a promising alternative that
grape stalks (25%), seeds (25%) and skins (50%) [4], and researches can achieve comparable oil yield with respect to the conventional
in the past few decades have shown that the possibilities of valoriz- organic solvent extraction with better product quality similar to
ing these by-products for the recovery of oil, phenolic compounds that of mechanical pressing.
and fibers are immense [5]. Comprehensive reviews appeared recently in the literature con-
Grape seeds oil is rich in unsaturated fatty acid and vitamin E cerning the SC-CO2 extraction technology and its perspective [12].
(tocopherols and tocotrienols) [6] and exhibits high antioxidant As a matter of fact, the SC-CO2 extraction process from solid sub-
activities which make it increasingly attractive in culinary, phar- strates is performed in the semi-continuous mode. The substrate,
maceutical, cosmetic and medical applications [7–9]. The oil yield in the form of a bed of particles, is stationary – contained in one or a
depends on extraction technique, type of solvent and operating series of extraction vessels – while the CO2 flows through it till the
solid is exhausted [13]. Designing such a kind of process requires,
among other things, selecting: the value of the process variables
∗ Corresponding author. Tel.: +39 0461 282692; fax: +39 0461 282672. (pressure, temperature, CO2 flow rate); to which extent the par-
E-mail address: (L. Fiori). ticles to be extracted have to be milled, i.e. the particle size; the
0896-8446/© 2015 Elsevier B.V. All rights reserved.
34 K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43

then the skins and seeds were separated by means of vibrating

Nomenclature sieves and further cleaned manually. Finally, the seeds were stored
in dark under vacuum at ambient temperature. Dried grape seeds
were milled by a grinder (Sunbeam Osterizer blender, Boca Raton,
a0 interfacial area (1/m)
USA) just before extraction. To avoid overheating, the sample was
dc oil bearing cell diameter (m)
flaked for 10 s, then grinding was halted and the sample was shook
dp particle diameter (m)
for another 10 s, and then the milling process was continued.
Dm binary diffusion coefficient (m2 /s)
Four representative grape cultivars, i.e. Muller Thurgau (MT),
D extractor diameter (m)
Pinot Noir (PI), Chardonnay (CH) and Moscato (MO), were selected
F solvent flow rate (g/min)
at random in this study. The oil yield for each cultivar was previ-
FM microstructural correction factor
ously measured, as well as the oil composition [6]. In particular,
G grinding efficiency
accounting for the great compositional similarities among the oils
kf external mass transfer coefficient (m/s)
from different grape cultivars [6], it was evaluated worth using
ks internal mass transfer coefficient (m/s)
grape seeds from different cultivars to achieve holistic results
i recursive index
which can be considered representative of any kind of grape seed
L extractor length (m)
oil. MT was used for evaluating the effect of pressure, temperature
n integer number
and solvent flow rate (Sections 4.1–4.3), PI was used to determine
Re Reynolds number
the effect of the particle size, bed porosity and extractor free vol-
 extractor free volume
ume (Sections 4.4, 4.5 and 4.7), and CH was used to study the effect
Vextr extractor volume (m3 )
of D/L (Section 4.6). MO was used to determine the grape seed oil
Vseeds volume occupied by the seeds (m3 )
solubility in SC-CO2 (Section 2.3).
x0 initial oil concentration in the solid (g oil/g seeds)
ys oil solubility in the solvent (g oil/g CO2 )
2.2. Supercritical CO2 extraction equipment and procedure
Greek Letters
˛ free oil correction factor In order to perform the extractions, the same equipment (PRO-
 exp extraction yield (experimental) (g oil/g seeds) RAS, Rome, Italy) was used with exactly the same procedure as
 model extraction yield (calculated by the model) detailed in [8]. The system was operated in the down-flow mode,
(g oil/g seeds) i.e. with the SC-CO2 flowing downwards through the substrate to be
f volume fraction of free oil extracted. Three different extractor baskets were utilized in the var-
f∗ adjusted volume fraction of free oil ious experimental runs, filled with appropriate mass of milled grape
seeds. The baskets consisted of hollow cylinders closed on both
ε extractor bed porosity
ends by metal frits. The frit at the top was intended to uniformly
distribute the solvent, while that at the bottom acted as structural
support for the solids and as filter medium. Fig. 1 reports the geom-
extractor diameter to length ratio (D/L); the compaction degree of etry of the extraction vessel (autoclave) and basket assembly. The
the bed of the milled particles – it is better to compact the bed, or
just to completely fill the extractor, or to leave some empty space
in the extractor?
This work analyses the effect of the above variables on the
extraction rate of oil from seeds, namely grape seeds. Even though
information on the effect of some operating conditions (pressure,
temperature, solvent flow rate and particle size) on the seed oil
extraction kinetics and yield is abundant [14–17], evidence of the
effect of parameters like D/L, bed porosity and bed free volume is
rather limited or completely missing in the literature.
For a more comprehensive analysis, an extraction model com-
monly adopted in the literature was utilized: the model by Sovová
[18]. This allowed to calculate, through best fitting procedures,
the key-parameters affecting the extraction kinetics (internal and
external mass transfer coefficients, “free oil” amount), and to eval-
uate how the process variables affect them. Moreover, this allowed
to compare and critically discuss the best fitted values of such key-
parameters based on well know and largely utilized correlations
available in the literature.
As a whole, the present paper provides information (trends
and punctual data) useful for addressing SC-CO2 extraction process

2. Materials and methods

2.1. Sample preparation

Grape marc was obtained from winemakers in Northern Italy.

Fig. 1. Extractor assembly: the various components of the three extractors. D
At the winery, stalks were separated from the seeds and skins. The
and L represent, respectively, the extraction basket internal diameter and length:
mixture of seeds and skins was taken to laboratory and stored at D = 4.07 × 10−2 m; L = 7.75 × 10−2 m (0.1 L basket), 15.5 × 10−2 m (0.2 L basket),
−20 ◦ C before drying. The samples were dried at 55 ◦ C for 48 h, and 38.3 × 10−2 m (0.5 L basket).
K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43 35

Fig. 2. Extraction curves at different pressures: oil yield versus solvent consumption. The operating conditions are reported in Table 1.

baskets had different internal volume, namely 0.1, 0.2 and 0.5 L. parameters for process design and insuring technical and economic
They had the same diameter, but different lengths (Fig. 1). When viability of the SC-CO2 extraction processes at industrial scale
using the 0.1 and 0.2 L baskets, tailor made spacers were used con- [22,23]. Numerous kinetics models for SC-CO2 extraction were pro-
sisting of stainless steel solid cylinders with a center hole to pipe posed in the literature to evaluate the extraction course. Among
(down-flow) the CO2 to the baskets. The utilization of the spacers them, it is worth mentioning the broken and intact cell (BIC) model
allowed to avoid the presence of empty spaces inside the auto- by Sovová [18,24], the shrinking core (SC) model by Goto et al. [25],
clave. The assembly was completed by a cap with a circular seal: the combined BIC-SC model by Fiori et al. [26,27]. Critical reviews
the basket, the (eventual) spacer and the cap were screwed together of the models are available in the literature [28–30].
and inserted into the autoclave (Fig. 1). The circular TeflonTM seal The models are mostly based on differential mass balances on
prevented from CO2 leakage. solid and fluid phases and differ from one another either by the
When analyzing the effect of pressure, temperature, solvent simplifying assumptions or the proposed mechanism of extrac-
flow rate, particle diameter and bed porosity, the 0.1 L extractor tion. In almost all the models, the mass balance equations are
basket was used. When analyzing the effect of D/L and extractor derived under general assumptions, such as: isothermal and iso-
free volume, all the three baskets were used. baric system, solvent free of solute at the extractor inlet, mono
Pressure, temperature and CO2 flow rate were kept con- pseudo-compound solute, constant bed porosity and constant
stant during the extraction process. The extraction operation was physical properties in the extractor, uniform distribution of solute
stopped when no more oil was extracted. After extractions, the par- in the solid and negligible axial dispersion.
ticle size distribution of the exhausted grape seeds was evaluated In this work the model proposed by Sovová [18] was applied,
by utilizing sieves having different mesh sizes placed in a vibrat- with the approximate analytical solution given by Št’astová et al.
ing device (Automatic Sieve Shaker D406 control, Auckland, New [31]. For details on the model, the reader can refer to the original
Zealand). The measure of the particle size distribution allowed to manuscripts [18,31]; the model equations and variables were also
calculate the Sauter mean diameter of the milled particle popula- recently reported [6].
tion, which was assumed as the reference value representative of The main parameters of the model are the following: the initial
the particle diameter, dp [19]. oil concentration in the solid, x0 ; the oil solubility in the solvent,
ys ; the bed void fraction, ε; the particle specific interfacial area, a0 ;
the external mass transfer coefficient, kf ; the internal mass transfer
2.3. Solubility determination
coefficient, ks ; the so-called grinding efficiency, G [31].
x0 and ys are input to the model: the former was previously
The solubility ys of grape seed oil at different temperatures and
calculated for the various cultivars [6], the latter was evaluated
pressures was determined by using a known mass of oil (5 g) previ-
here according to the procedure reported in Section 2.3.
ously extracted by SC-CO2 . The oil was dispersed on the surface of
ε can be easily calculated considering the mass and the density
glass beads placed inside the 0.1 L basket, and then re-extracted by
of seeds charged (1103 kg/m3 [8]), and the extractor basket volume.
SC-CO2 at low solvent flow rate (about 4 g CO2 /min). According to
The other variables represent the adjustable parameters of
the dynamic method for measuring the solubility in supercritical
the model. When utilizing the model in best fitting experimental
solvent [20], the initial slope of the extraction curve was used to
extraction curves, the optimization routine provides the optimum
calculate the solubility at the given pressure and temperature. As
value of the adjustable parameters. In particular, for each experi-
mentioned in Section 2.1, the oil solubility values were obtained
mental extraction curve, the values of kf a0 , ks a0 and G are obtained.
utilizing MO, but the values are representative for all the cultivars,
The model was written as a MATLABTM code to best fit the exper-
as grape seed oil composition is extremely similar for the different
imental data and to obtain the value of the adjustable parameters.
cultivars [6]. In addition, the solubility values here determined are
The goodness of the model fitting was evaluated and quantified
consistent with previously measured ones [21].
calculating, for each experimental run, the percent average abso-
lute relative deviation (AARD (%)), given by Eq. (1), and the root
3. Mathematical modeling mean square error (RMSE), calculated according to Eq. (2).
1  (exp − model ) 
The evaluation of the overall extraction curves through kinetics
modeling is of paramount importance for obtaining the opti- AARD(%) =   100 (1)
n exp
mum operating conditions, scaling-up the process, determining i=1
36 K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43

Table 1
Operating conditions and estimated model adjustable parameters for different pressures (T = 40 ◦ C, ε = 0.41, x0 = 0.120).

P (bar) ys (mg/g) dp (mm) F (g/min) G kf a0 (s−1 ) × 102 ks a0 (s−1 ) × 105 RMSE × 102 AARD (%)

200 4.20 0.39 8.51 0.71 1.255 3.49 0.871 3.24

300 7.60 0.41 8.43 0.76 0.980 3.45 0.467 1.77
400 10.4 0.48 8.32 0.72 0.829 7.18 0.352 3.31
500 13.0 0.37 8.59 0.72 0.661 9.70 0.375 2.08

 n operating variables and model adjustable parameters with asso-
 (exp − model )2
RMSE =  (2) ciated deviation from experimental data represented in terms of
n RMSE and AARD.
The trend is that expected. Keeping constant the mass of solid
where n represents the number of available experimental data, and in the extractor (namely 65 g), the time necessary to complete the
␥exp and ␥model are, respectively, the experimental extraction yield linear section of extraction curve reduced from 270 to 55 min when
and the extraction yield predicted by the model – yield expressed the pressure increased from 200 to 500 bar. The increase in extrac-
as mass of extracted oil per mass of seeds. tion rate reflected the increase in oil solubility ys , as shown in
Table 1.
4. Results and discussion Similar values of the adjustable parameters (G, kf a0 and ks a0 )
were obtained in the whole range of pressure investigated. More
The results of experimental tests and modeling are reported in precisely, increasing the pressure made kf a0 slightly to decrease
the present section. From Sections 4.1–4.7, the effect of each sin- and ks a0 slightly to increase. The behavior of kf a0 reported in Table 1
gle process variable is analyzed separately, first referring to the is consistent with the dependence of kf on Schmidt and Reynolds
experimental outcomes, then to the output from model best fitting. numbers [32] (see Section 4.8 for details).
Section 4.8 presents as a whole the best fitted values of the mod- The values of ks a0 in Table 1 can be due, to some extent, to
els parameter, and analyzes them resorting to correlations largely the slightly different value of the particle diameter of the different
utilized in the literature. trials. The internal mass transfer coefficient is inversely propor-
tional to the particle diameter: Table 1 confirms this trend with
4.1. Effect of pressure one exception, at P = 400 bar.
Comparable results and similar trends of the model parameters
The effect of SC-CO2 pressure on kinetics of extraction is well were reported for the SC-CO2 extraction of apricot kernel oil by
established, rather solid and there is a consensus in the research Özkal et al. [16].
community that increasing operating pressure has a positive effect
on the extraction rate. The reason is that an increase in pressure 4.2. Effect of temperature
(at constant temperature) makes the density of SC-CO2 increase,
which enhances its solvent power and, ultimately, the extraction The effect of SC-CO2 temperature on the extraction kinetics
rate if all the other parameters are kept constant. Nevertheless, the is rather conflicting as a result of what is known as “crossover
economic feasibility of working at elevated pressure has to be eval- phenomena”. When temperature increases, the density of SC-CO2
uated on a case-by-case, as any increase in pressure is associated decreases, but the solute solubility can still increase as a result of
with an increase in energy consumption. In the case of oil from enhanced solute vapor pressure. The plots of solubility versus pres-
seeds, working at high pressure seems economically convenient sure at constant but different temperatures cross each other twice
[13]. and these intersections are referred as lower and upper crossover
In this work the pressure was varied in the range 200–500 bar, at pressure points [33]. At pressures between these two points, solu-
a constant temperature of 40 ◦ C, with CO2 flow rate at 8.46 ± 0.12 g bility decreases with increase in temperature because the solvent
CO2 /min, particle diameter of 0.41 ± 0.05 mm and constant bed density effect overcomes the vapor pressure effect. Whereas above
porosity of 0.41. The extraction kinetics is shown in Fig. 2, where the the upper or below the lower crossover point the vapor pressure
extraction yield is reported versus CO2 consumption. Table 1 reports effect is more pronounced than the density effect, so the solubility
the characteristic values of each experimental trial: the value of increases with an increase in temperature [33–35].

Fig. 3. Extraction curves at different temperatures: oil yield versus solvent consumption. The operating conditions are reported in Table 2.
K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43 37

Table 2
Operating conditions and estimated model adjustable parameters for different temperatures (P = 500 bar, ε = 0.41, x0 = 0.120).

T (◦ C) ys (mg/g) dp (mm) F (g/min) G kf a0 (s−1 ) × 102 ks a0 (s−1 ) × 105 RMSE × 102 AARD (%)

35 12.8 0.41 8.28 0.70 0.302 6.44 0.444 1.88

40 13.0 0.37 8.59 0.72 0.661 9.70 0.375 2.08
50 13.4 0.32 8.70 0.77 0.506 6.19 0.292 1.29

Fig. 4. Extraction curves at different solvent flow rates. (a) oil yield versus solvent consumption; (b) oil yield versus time. The operating conditions are reported in Table 3.

In this work, the effect of temperature was studied at a pres- In this case both kf a0 and ks a0 present similar values for all the
sure of 500 bar, which is above the upper crossover point. The tests, and no specific trend can be identified.
other extraction conditions were as follows: flow rate of 8.52 ± 0.2 g
CO2 /min, particle size of 0.37 ± 0.04 mm, temperatures of 35, 40
and 50 ◦ C. 4.3. Effect of flow rate
Fig. 3 shows the extraction kinetics and Table 2 reports the char-
acteristic value of the various parameters of each experimental test. The effect of flow rate was studied at four different conditions,
The extraction rate slightly increased with an increase in temper- namely 4.71, 7.45, 8.43 and 10.22 g CO2 /min. The other extraction
ature, and so did the solubility. The difference in value of the final conditions were as follows: pressure of 350 bar, temperature of
asymptotic oil yields is very likely not related to temperature, but 40 ◦ C, particle size of 0.42 ± 0.01 mm.
it is rather due to the slight difference in particle diameter, which The experiments were designed with the following approach.
is also manifested in the value of the grinding efficiency G. Refer to At first, the test at the highest flow rate (10.22 g CO2 /min) was per-
Section 4.4 for an in deep discussion about the dependence of the formed, and in the subsequent tests the flow was reduced till a flow
extraction kinetics on the particle size. rate value (4.71 g CO2 /min) at which the slope of the extraction

Table 3
Operating conditions and estimated model adjustable parameters for different flow rates (T = 40 ◦ C, P = 350 bar, ys = 8.60 mg/g, ε = 0.41, x0 = 0.120).

dp (mm) F (g/min) G kf a0 (s−1 ) × 102 ks a0 (s−1 ) × 105 RMSE × 102 AARD (%)

4.71 0.42 0.52 0.384 2.30 0.497 1.95

7.45 0.43 0.62 0.698 5.04 0.407 1.51
8.43 0.41 0.57 1.002 8.87 0.476 1.87
10.22 0.43 0.78 1.270 9.73 0.573 3.93
38 K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43

Fig. 5. Extraction curves at different particle diameters: oil yield versus solvent consumption. The operating conditions are reported in Table 4.

Table 4
Operating conditions and estimated model adjustable parameters for different particle sizes (T = 50 ◦ C, P = 500 bar, ys = 13.4 mg/g, ε = 0.41, x0 = 0.167).

dp (mm) F (g/min) G kf a0 (s−1 ) × 102 ks a0 (s−1 ) × 105 RMSE × 102 AARD (%)

0.41 7.34 0.81 0.326 4.98 0.490 1.49

0.45 7.19 0.67 0.427 2.44 1.140 4.78
0.59 7.46 0.55 0.242 2.56 0.993 3.92
0.75 7.31 0.39 0.611 1.98 0.483 2.80

Table 5
Operating conditions and estimated model adjustable parameters for different bed porosity (T = 50 ◦ C, P = 500 bar, ys = 13.4 mg/g, x0 = 0.167).

␧ dp (mm) F (g/min) G kf a0 (s−1 ) × 102 ks a0 (s−1 ) × 105 RMSE × 102 AARD (%)

0.41 0.38 8.84 0.81 0.633 9.63 0.417 2.31

0.32 0.47 8.38 0.72 0.487 11.4 1.129 5.39
0.23 0.43 8.63 0.86 0.549 10.8 0.451 2.02
0.10 0.47 8.43 0.93 0.177 2.33 0.974 1.44

curve coincided with the solubility determined using glass beads The results reported in Table 4 are fully consistent with the above
(ys = 8.60 mg/g at 350 bar and 40 ◦ C). statements: ks a0 and the grinding efficiency G gradually decrease
Fig. 4a reports the oil yield versus CO2 consumption. Fig. 4b with the increase in particle size. Fig. 5 clearly shows that the initial
reports the oil yield versus time. The higher was the flow rate, slope of the extraction curves overlap, which can also be verified
the higher the extraction rate (Fig. 4b), in line with an increase in from the values of the external mass transfer parameter (kf a0 ) of
both the external and internal mass transfer parameters (Table 3). Table 4: the values are in the same order of magnitude without any
While the increase in kf a0 at increasing flow rate was expected, specific trend.
the increase in ks a0 at increasing flow rate is difficult to explain, as
the particle diameters were very similar for the various tests and 4.5. Effect of bed porosity
ks should not depend on flow rate on a theoretical basis. Anyway,
such kind of anomalous dependence was already previously found The effect of the bed porosity (or, equivalently, bed void fraction)
in the literature [16,36]. Thus, the higher was the flow rate, the on the extraction kinetics was evaluated with bed void fraction in
lower the extraction time. But, conversely, the specific consump- the range 10–41%. The other extraction conditions were as follows:
tion of the solvent increased at increasing SC-CO2 flow rate (Fig. 4a). pressure of 500 bar, temperature of 50◦ C, flow rate of 8.60 ± 0.23 g
Ultimately, for commercial applications the solvent flow rate has to CO2 /min, particle diameter of 0.44 ± 0.04 mm. In all the cases the
be optimized in terms of extraction time and solvent volume used extractor basket was full, but compacted to a different degree. In
per operation. the case of ε = 0.41, the bed of particles was not compacted at all.
Consequently, each test was characterized by a different amount of
4.4. Effect of particle diameter substrate charged into the extractor basket.
The extraction kinetic curves are presented in Fig. 6a (oil yield
Fig. 5 shows the extraction kinetic curves for four different par- versus CO2 consumption) and Fig. 6b (oil extracted versus time).
ticle diameters, namely: 0.41, 0.45, 0.59 and 0.75 mm. The other The choice of reporting in the y-axis of Fig. 6b the oil (instead than
extraction conditions were as follows: flow rate of 7.33 ± 0.10 g the oil yield) allows for a correct quantification of the oil extraction
CO2 /min, pressure of 500 bar, temperature of 50◦ C. Fig. 5 testi- rate.
fies that the asymptotic oil yield decreased with the increase in It is remarkable that, in the initial stage of the process, the
particle size. Fine particles are easier to extract because they have extraction rate did not depend on the bed porosity when this
large surface area per unit volume, contain a high percentage of parameter was in range 0.23–0.41 (Fig. 6b). A further decrease of
“free oil” and require less distance for the “tied oil” to reach the the bed porosity to 0.10 had a negative effect on the extraction rate,
surface, which reduces the internal mass transfer resistance [18]. which was probably due to flow inhomogeneity (channeling) due to
K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43 39

Fig. 6. Extraction curves at different particle bed porosity. (a) oil yield versus solvent consumption; (b) oil extracted versus time. The operating conditions are reported in
Table 5.

the high compaction degree. Another potential cause is the reduced diameter but different lengths were utilized. To maintain constant
residence time of the solvent into the extractor at the reduced bed the bed porosity (ε = 0.41) in all the three extractors, 65, 130 and
porosity: the CO2 residence time decreased from about 270 s at 325 g of solid matrix were charged, respectively, in the basket of 0.1,
ε = 0.41 to about 70 s at ε = 0.10. 0.2 and 0.5 L volume. To preserve a constant solvent residence time,
The values of the external and internal mass transfer parameters the ratio of mass of substrate to CO2 flow rate was kept constant
are quite similar for bed porosity in the range 0.23–0.41 (Table 5). roughly at 10 g seeds/g CO2 /min. This approach is often selected as
Conversely, the value of both parameters drops down at bed poros- scale-up criterion in SC-CO2 extraction process development [37].
ity equal to 0.10. Vice-versa, the value of the grinding efficiency G The values of solvent flow rate are reported in Table 6: correspond-
is very large for ε = 0.10. This was rather unexpected considering ingly, the superficial SC-CO2 velocity increased from 0.084 mm/s for
that G should reflect the particle diameter, which is quite similar the 0.1 L basket to 0.45 mm/s for the 0.5 L basket. The other extrac-
for all the tests this section refers to. A possible explanation could tion conditions were as follows: pressure of 350 bar, temperature
be that the best fitting procedure has to cope with an extraction of 40 ◦ C, particle diameter of 0.42 ± 0.04 mm.
initial slope (too) much lower than the oil solubility. The only way The extraction rate increased when D/L decreased, i.e. the longer
the model can fit such a curve (low initial slope) is to assume a very was the extractor basket, the lower the specific solvent consump-
low value for the external mass transfer parameter (kf a0 ). In the fol- tion: Fig. 7. Importantly, in these experimental runs it is believed
lowing extraction stage, the model compensates by a large value of that the solvent was not fully saturated by the solute. According to
G. If this was the case, the goodness of the value of the adjustable the results of Section 4.3, for grape seeds the saturation occurred
parameters reported in Table 5 for ε = 0.10 would go beyond their at values of about 14 g seeds/g CO2 /min.
physical meaning, and therefore these values should be considered The differences in extraction rate when varying D/L are reflected
with caution. in the values of the external mass transfer parameter: the lower was
D/L, the higher kf a0 (Table 6). This result is clearly consistent with
4.6. Effect of extractor diameter to length ratio (D/L) the dependence of kf on solvent flow rate.
The values of the grinding efficiency G are consistent with the
Fig. 7 shows the effect of D/L on the extraction kinetics. As men- small differences in the particle diameters. No specific trend can be
tioned in Section 2.2, three extraction baskets having the same observed for ks a0 .
40 K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43

Fig. 7. Extraction curves at different extractor diameter to length ratios: oil yield versus solvent consumption. The operating conditions are reported in Table 6.

Table 6
Operating conditions and estimated model adjustable parameters for different D/L (T = 40 ◦ C, P = 350 bar, ys = 83.60 mg/g, x0 = 0.147).

D/L dp (mm) F (g/min) G kf a0 (s−1 ) × 102 ks a0 (s−1 ) × 105 RMSE × 102 AARD (%)

0.53 0.47 6.11 0.66 0.679 3.03 0.998 2.97

0.26 0.40 12.97 0.68 0.871 2.47 0.748 6.13
0.11 0.38 32.78 0.75 1.009 4.95 0.594 3.16

4.7. Extractor free volume Conversely, the extractor basket of 0.2 L was roughly half full, and
that of 0.5 L was empty for more than three quarters.
The extractor free volume  is different from the extractor In the case of the 0.2 L ( = 55%) and 0.5 L ( = 82%) extractors,
bed porosity ε discussed in Section 4.5. In this case, there can be the extraction curves mostly overlap (Fig. 8). But, when analyzing
an empty space above the solid matrix in the extractor basket. Table 7, the corresponding values of the external and internal mass
The extractor free volume accounts also for this empty space. The transfer parameters are unexpectedly different. One possible expla-
extractor free volume was thus computed as the percentage of nation is that the model was utilized outside its validity range. In
empty space in the extractor: other words, reasonably is not physically consistent to let the value
of the bed porosity (model variable) equal to the extractor free vol-
(Vextr − Vseeds ) ume calculated according to Eq. (3). This is most likely the case
 = 100 (3)
Vextr considering, in addition, that it is not known what is happening
inside the extractor basket under these conditions. Grape seed par-
The same mass of grape seeds (100 g) was utilized inside the ticles have a mean density greater (but not so dissimilar) than that
three different extraction baskets. In the case of the smallest extrac- of SC-CO2 and the extractor is operated in the down-flow mode.
tor, the test is actually the same already utilized in the discussion This would suggest that the bed of particles stays stable at the bot-
of Section 4.5 concerning bed porosity; i.e. for Vextr = 0.1 L, the indi- tom of the basket. Nevertheless, it cannot be a priori excluded that
cation of a free volume  = 10% (Fig. 8) coincides with ε = 0.10 the bed of particles (or a portion of it) spreads along the entire vol-
(Fig. 6a and b). In that case the matrix completely filled the basket. ume of the extractor basket. In addition, it is not known is the plug

Fig. 8. Extraction curves at different extractor free volume: oil yield versus solvent consumption. The operating conditions are reported in Table 7.
K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43 41

Table 7
Operating conditions and estimated model adjustable parameters for different extractor free volume (T = 50 ◦ C, P = 500 bar, ys = 13.4 mg/g, mseeds = 100 g, x0 = 0.167).

 (%) dp (mm) F (g/min) G kf a0 (s−1 ) × 102 ks a0 (s−1 ) × 105 RMSE × 102 AARD (%)

10 0.47 8.43 0.93 0.177 2.33 0.974 1.44

55 0.44 8.87 0.69 0.466 8.37 0.489 1.21
82 0.44 8.35 0.62 0.243 3.52 1.244 2.85

Fig. 9. Free oil amount versus particle diameter. Gx0 : g free oil/g seeds; f and f∗ : cm3 free oil/cm3 seed particle. Filled circles: significant data. Empty circles: questionable
data from Table 7.

flow assumption is still applicable. To get a clearer picture, further where dc is the diameter of the oil bearing cell (dc = 20 ␮m for grape
investigation is required. At this stage, the (quite random) values of seeds [26,39]).
the model adjustable parameters of Table 7 should be considered Fig. 9 reports the values of Gx0 , f and f∗ as a function of the
with caution, while the experimental trends of Fig. 8 maintain their mean particle diameter. Gx0 values (represented as circles in Fig. 9)
scientific interest. were calculated from the values of Tables 1–7.
The questionable results of Table 7 were represented as empty
4.8. Critical evaluation of the key-parameters affecting extraction circles in Fig. 9.
kinetics Gx0 values show the expected decreasing trend as dp increases.
Even if they exhibit not negligible scattering, it is worth noticing
This section addresses the modeling output from best fitting of
that they locate between curves f and f∗ , testifying that the free
experimental data, i.e. the values already presented in Tables 1–7
which have been here appropriately rearranged: namely, the oil content by the BIC model of Sovovà [18] is comparable to the
amount of free oil and the external and internal mass transfer free oil by the BIC model by Reverchon and Marrone [38,39] and its
coefficients. modification by Fiori and Costa [40]. Importantly, similar amount of
The mass fraction of free oil (g free oil/g seeds) is given by Gx0 free oil results also from the BIC-SC models by Fiori when adopting
[31]. Gx0 can be compared to the volume fraction of free oil f (cm3 the double shell hypothesis [26,27], as it was previously discussed
free oil/cm3 seed particle) as proposed by Reverchon and Marrone [40].
[38,39]: The values of the external and internal mass transfer coefficients
can be computed from the values reported in Tables 1–7 and resort-
dc ing to Eq. (6) to estimate a0 :
f = 3 (4)
6(1 − ε)
and to the value of such variable as later modified by Fiori and Costa a0 = (6)
[40] by introducing a free oil correction factor ˛ which takes into
account the amount of oil characteristic of the vegetable species The kf values can be compared with the values predicted using
(˛ = 0.472 for grape seeds [40]): various literature correlations valid for packed beds operating with
supercritical fluids. Fig. 10 shows, on a parity plot, the kf values
f∗ = 3 ˛ (5) here obtained (kfMod ) and the kf values obtained using the cor-
dp relation proposed by Mongkholkhajornsilp et al. [41] (kfMDDETP ).
42 K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43

even if a little smaller, are in reasonably agreement with theoretical

Theoretically, the internal mass transfer coefficient should not
depend on SC-CO2 flow rate and bed porosity, while it should
depend (to a small extent) on pressure and temperature and (to
a large extent) on particle diameter. In an attempt to take account
of these dependencies, del Valle et al. [45] proposed the utilization
of the dimensionless number referred as microstructural correc-
tion factor FM which, after simple rearrangements, can be written
ks dp
FM = (7)
Fig. 11 reports FM calculated according to Eq. (7) versus the
particle diameter. Values range between 2 × 10−4 and 1.5 × 10−3 ,
varying by almost one order of magnitude. Referring to the data
obtained when studying the effect of the particle size (i.e. those
deducted from Table 4 and indicated in Fig. 11 with the sym-
bol “star”) the scatter reduces significantly. On the one hand, this
indicates that Eq. (7) manages to handle quite satisfactorily the
dependence of ks on dp ; on the other hand, it is self-evident that
the ks data here obtained suffer from an intrinsic scatter which goes
beyond the theoretical dependence of ks on the operating variables.
Fig. 10. Comparison between the external mass transfer coefficient by this work
(kfMod ) and the external mass transfer coefficient by the correlation proposed by 5. Conclusions
Mongkholkhajornsilp et al. [41] (kfMDDETP ). Filled circles: significant data. Empty
circles: questionable data from Table 7.
The effect of the main process variables affecting the super-
critical CO2 extraction of oil from seeds (namely grape seeds) was
Such a correlation was chosen considering that it applies to low investigated, both experimentally and trough modeling.
Reynolds number (Re) like those characterizing the tests performed The extraction rate increased with an increase in pressure, tem-
(0.25 ≤ Re ≤ 0.66 for all the tests but one where Re = 1.59). To cal- perature and solvent flow rate (but in this case the CO2 specific
culate kfMDDETP , the density and viscosity of the supercritical fluid consumption also increased). At a fixed ratio of mass of seeds to
were assumed as those of pure CO2 as available in the NIST database solvent flow rate, decreasing the extractor diameter to length ratio
[42]. The value of the binary diffusion coefficient Dm of oil in SC- allowed to reduce the CO2 specific consumption. An extractor bed
CO2 was estimated using the correlation by Catchpole and King porosity in the range of 0.23–0.41 had no effect of the extraction
[43], simplifying the oil as consisting of triolein. kinetics, but a further decrease in bed porosity resulted detrimen-
Fig. 10 testifies that the kf values here obtained are up to an tal, probably due to the occurring of channeling. The particle size
order of magnitude lower than the theoretical values: in most cases had no effect on the initial extraction rate, but reflected on the
0.2kfMDDETP < kfMod < 0.5kfMDDETP . For further comparison, theoreti- final asymptotic extraction yield, the smaller were the particles,
cal values of the external mass transfer coefficient were calculated the higher the final yield.
using the correlation of Puiggene et al. [44] (kfPLR ), which is actually The experimental extraction data were modeled through the BIC
valid for higher Re (10 < Re < 100). Similar results were also obtained model by Sovová [18]. Through best fitting procedures, the inter-
in this case: 0.1kfPLR < kfMod < 0.5kf PLR. To summarize, it is possible nal and external mass transfer parameters as well as the free oil
to affirm that the kf values computed in the present investigation, content were calculated: in most cases their dependence on the

Fig. 11. FM = ks dp /Dm versus particle diameter. Filled circles: significant data. Empty circles: questionable data from Table 7. “Star” symbols: data relevant to Table 4.
K.S. Duba, L. Fiori / J. of Supercritical Fluids 98 (2015) 33–43 43

process variables was as expected. The best fitted values of the [18] H. Sovová, Rate of the vegetable oil extraction with supercritical CO2 -I. model-
model parameters were discussed based on literature correlations ing of extraction curves, Chemical Engineering Science 49 (1994) 409–414.
[19] L. Fiori, D. Basso, P. Costa, Seed oil supercritical extraction: particle size dis-
and other extraction models available in the literature. This allowed tribution of the milled seeds and modeling, J. Supercritical Fluids 47 (2008)
a critical comparison showing that the values of the external mass 174–181.
transfer coefficient here obtained are comparable but slightly lower [20] G. Brunner, Gas Extraction: An Introduction to Fundamentals of Supercriti-
cal Fluids and the Application to Separation Processes, Darmstadt: Steinkopff:
than those theoretically predicted, while the amount of free oil Springer, New York, 1994, pp. 125–127.
is in very good agreement with that foreseen by other extraction [21] H. Sovová, M. Zarevúcka, M. Vacek, K. Stránský, Solubility of two vegetable oils
models. Finally, the values of the internal mass transfer coefficient in supercritical CO2 , J. Supercritical Fluids 20 (2001) 15–28.
[22] O. Döker, U. Salgın, İ. Şanal, Ü. Mehmetoğlu, A. Çalımlı, Modeling of extrac-
(rearranged in a dimensionless form) resulted scattered by almost
tion of ␤-carotene from apricot bagasse using supercritical CO2 in packed bed
one order of magnitude. extractor, J. Supercritical Fluids 28 (2004) 11–19.
The BIC model allowed for a very good fitting of the experimental [23] S. Machmudah, A. Sulaswatty, M. Sasaki, M. Goto, T. Hirose, Supercritical CO2
data, with maximum root mean square error of 1.20 × 10−2 and extraction of nutmeg oil: experiments and modeling, J. Supercritical Fluids 39
(2006) 30–39.
percent average absolute relative deviation of 6.1% considering all [24] H. Sovová, Steps of supercritical fluid extraction of natural products and their
the investigated conditions. characteristic times, J. Supercritical Fluids 66 (2012) 73–79.
[25] M. Goto, B.C. Roy, T. Hirose, Shrinking-core leaching model for supercritical-
fluid extraction, J. Supercritical Fluids 9 (1996) 128–133.
Acknowledgments [26] L. Fiori, D. Basso, P. Costa, Supercritical extraction kinetics of seed oil: a new
model bridging the “broken and intact cells” and the “shrinking-core” models,
Research supported by AGER (project number 2010-2222). J. Supercritical Fluids 48 (2009) 131–138.
[27] L. Fiori, Supercritical extraction of sunflower seed oil: experimental data and
model validation, J. Supercritical Fluids 50 (2009) 218–224.
References [28] J.M. del Valle, J.C. de la Fuente, Supercritical CO2 extraction of oilseeds: review of
kinetic and equilibrium models, Critical Reviews in Food Science and Nutrition
[1] F. Castellucci, B. Lasiello, Statistical Report on World Vitiviniculture, Paris, 2013, 46 (2006) 131–160.
Available from: [29] E.L.G. Oliveira, A.J.D. Silvestre, C.M. Silva, Review of kinetic models for super-
[2] C. Sloop, O. Bettini, EU-27 Wine Annual Report and Statistics, Washing- critical fluid extraction, Chemical Engineering Research and Design 89 (2011)
ton D.C., 2013, Available from: 1104–1117.
default.asp [30] Z. Huang, X.-H. Shi, W.-J. Jiang, Theoretical models for supercritical fluid extrac-
[3] T. Bacic, Recovery of Valuable Products from Lees and Integrated Approach to tion, J. Chromatography A 1250 (2012) 2–26.
Minimise Waste and Add Value to Wine Production, Melbourne, 2003, Available [31] J. Št’astová, J. Jež, M. Bártlová, H. Sovová, Rate of the vegetable oil extraction
from: with supercritical CO2 -III. Extraction from sea buckthorn, Chemical Engineer-
[4] B.J. Kelly, M. Joan, Wine industry residues, in: P.S. nee’ Nigam, A. Pandey (Eds.), ing Science 51 (1996) 4347–4352.
Biotechnology for Agro-Industrial Residues Utililisation, 1st ed., Springer, The [32] L. Fiori, D. Calcagno, P. Costa, Sensitivity analysis and operative conditions of a
Netherlands, 2009, pp. 293–311. supercritical fluid extractor, J. Supercritical Fluids 41 (2007) 31–42.
[5] A.J. Shrikhande, Wine by-products with health benefits, Food Research Inter- [33] S.A.B. Vieira de Melo, G.M.N. Costa, A.C.C. Viana, F.L.P. Pessoa, Solid pure compo-
national 33 (2000) 469–474. nent property effects on modeling upper crossover pressure for supercritical
[6] L. Fiori, V. Lavelli, K.S. Duba, P.S.C. Sri Harsha, H.B. Mohamed, G. Guella, Super- fluid process synthesis: a case study for the separation of Annatto pigments
critical CO2 extraction of oil from seeds of six grape cultivars: modeling of using SC-CO2 , J. Supercritical Fluids 49 (2009) 1–8.
mass transfer kinetics and evaluation of lipid profiles and tocol contents, J. [34] S.G. Özkal, U. Salgın, M.E. Yener, Supercritical carbon dioxide extraction of
Supercritical Fluids 94 (2014) 71–80. hazelnut oil, J. Food Engineering 69 (2005) 217–223.
[7] L. Fernandes, S. Casal, R. Cruz, J.A. Pereira, E. Ramalhosa, Seed oils of ten tra- [35] G. Anitescu, L.L. Tavlarides, Solubility of individual polychlorinated biphenyl
ditional Portuguese grape varieties with interesting chemical and antioxidant (PCB) congeners in supercritical fluids: CO2 , CO2 /MeOH and CO2 /n-C4 H10 , J.
properties, Food Research International 50 (2013) 161–166. Supercritical Fluids 14 (1999) 197–211.
[8] L. Fiori, Grape seed oil supercritical extraction kinetic and solubility data: crit- [36] C. Grosso, J.P. Coelho, F.L.P. Pessoa, J.M.N.A. Fareleira, J.G. Barroso, J.S. Urieta,
ical approach and modeling, J. Supercritical Fluids 43 (2007) 43–54. A.F. Palavra, H. Sovová, Mathematical modelling of supercritical CO2 extraction
[9] C.P. Passos, R.M. Silva, F.A. Da Silva, M.A. Coimbra, C.M. Silva, Enhancement of volatile oils from aromatic plants, Chemical Engineering Science 65 (2010)
of the supercritical fluid extraction of grape seed oil by using enzymatically 3579–3590.
pre-treated seed, J. Supercritical Fluids 48 (2009) 225–229. [37] M.M.R. de Melo, R.M.A. Domingues, M. Sova, E. Lack, H. Seidlitz, F. Lang Jr., A.J.D.
[10] C. Da Porto, E. Porretto, D. Decorti, Comparison of ultrasound-assisted extrac- Silvestre, C.M. Silva, Scale-up studies of the supercritical fluid extraction of
tion with conventional extraction methods of oil and polyphenols from grape triterpenic acids from Eucalyptus globulus bark, J. Supercritical Fluids 95 (2014)
(Vitis vinifera L.) seeds, Ultrasonic Sonochemistry 20 (2013) 1076–1080. 44–50.
[11] L.D.S. Freitas, C. Dariva, R.A. Jacques, E.B. Caramão, Effect of experimental [38] C. Marrone, M. Poletto, E. Reverchon, A. Stassi, Almond oil extraction by super-
parameters in the pressurized liquid extraction of brazilian grape seed oil, critical CO2 : experiments and modelling, Chemical Engineering Science 53
Separation and Purification Technology 116 (2013) 313–318. (1998) 3711–3718.
[12] M.M.R. de Melo, A.J.D. Silvestre, C.M. Silva, Supercritical fluid extraction of veg- [39] E. Reverchon, C. Marrone, Modeling and simulation of the supercritical CO2
etable matrices: applications, trends and future perspectives of a convincing extraction of vegetable oils, J. Supercritical Fluids 19 (2001) 161–175.
green technology, J. Supercritical Fluids 92 (2014) 115–176. [40] L. Fiori, P. Costa, Supercritical extraction of seed oil: analysis and comparison
[13] L. Fiori, Supercritical extraction of grape seed oil at industrial-scale: plant of up-to-date models, in: M.R. Belinsky (Ed.), Supercritical Fluids, Nova Science
and process design, modeling, economic feasibility, Chemical Engineering and Publishers, Inc., New York, 2011, pp. 705–722.
Processing: Process Intensification 49 (2010) 866–872. [41] D. Mongkholkhajornsilp, S. Douglas, P.L. Douglas, A. Elkamel, W. Teppaitoon,
[14] U. Salgın, S. Salgın, Effect of main process parameters on extraction of pine ker- S. Pongamphai, Supercritical CO2 extraction of nimbin from neem seeds – a
nel lipid using supercritical green solvents: solubility models and lipid profiles, modelling study, J. Food Engineering 71 (2005) 331–340.
J. Supercritical Fluids 73 (2013) 18–27. [42] NIST, Fluid Thermodynamic and Transport Properties, Version 5.0: Available
[15] H. Mhemdi, E. Rodier, N. Kechaou, J. Fages, A supercritical tuneable process for from:
the selective extraction of fats and essential oil from coriander seeds, J. Food [43] O.J. Catchpole, M.B. King, Measurement and correlation of binary diffusion
Engineering 105 (2011) 609–616. coefficients in near critical fluids, Industrial & Engineering Chemistry Research
[16] S.G. Özkal, M.E. Yener, L. Bayındırlı, Mass transfer modeling of apricot kernel oil 33 (1994) 1828–1837.
extraction with supercritical carbon dioxide, J. Supercritical Fluids 35 (2005) [44] J. Puiggené, M.A. Larrayoz, F. Recasens, Free liquid-to-supercritical fluid mass
119–127. transfer in packed beds, Chemical Engineering Science 52 (1997) 195–212.
[17] C.M. Fernández, L. Fiori, M.J. Ramos, Á. Pérez, J.F. Rodríguez, Supercritical [45] J.M. del Valle, J.C. Germain, E. Uquiche, C. Zetzl, G. Brunner, Microstructural
extraction and fractionation of Jatropha curcas L. oil for biodiesel production, J. effects on internal mass transfer of lipids in prepressed and flaked vegetable
Supercritical Fluids 97 (2015) 100–106. substrates, J. Supercritical Fluids 37 (2006) 178–190.