Beruflich Dokumente
Kultur Dokumente
a r t i c l e i n f o a b s t r a c t
Article history: Based on the BrunauerEmmettTeller (BET) theory of adsorption, two mathematical models for cor-
Received 6 May 2014 relating supercritical extraction data were developed. Each one of the models expresses the extraction
Received in revised form yield as a function of time through one simple equation, which contains only three or two adjustable
26 September 2014
parameters, respectively, each one with a clear physical meaning. A comparison of the performance of
Accepted 27 September 2014
the new models with those of Sovovs and VT-II models shows that the rst proposed model, with three
Available online 13 October 2014
adjustable parameters, produces in many cases an average deviation that is similar to that produced by
the Sovovs model, and is lower than that by the VT-II model, a remarkable result given the simplicity of
Keywords:
Supercritical extraction
the model. In addition, the second proposed model, which only has two adjustable parameters, performs
Mathematical models well as compared to the rigorous models especially in the rst period of the extraction and can be useful
Extraction kinetics for generating rst estimates of the parameters for data correlation with the rst model. These results
Adsorption indicate that the proposed models are simple, useful and quick tools for correlating experimental data
BET on supercritical extraction of natural matters.
Natural matters 2014 Elsevier B.V. All rights reserved.
Mathematical models for correlating experimental data on 2.1. Material balance in the bulk uid phase
supercritical extraction of natural matter are important tools for
scaling laboratory data up to industrial level, in order to make esti- Based on the general model for the supercritical uid (SCF)
mates of protability or to design processing equipment. Many extraction of a solid substrate in a packed bed [5], we initially con-
models have thus been proposed, as referenced in comprehensive sider a packed extractor, shown in Fig. 1, and write the material
literature reviews that have been published [16]. Among them, balance for the solute in the supercritical uid, around a differential
that by Sovov [1] and the so-called VT-II [2] are examples of rig- element along the axial direction of the extractor:
orous models. Sovovs model is based on the representation of
diffusive and convective transport phenomena that occur during Cf Cf 2 Cf (1 )
+ u = Dsf + asf ksf Cf Cf (1)
the extraction and has three analytical equations, one for each t z z 2
extraction period. The VT-II model represents the same phenomena
where Cf is the concentration of the solute in the SCF, u is the inter-
through a set of differential equations that has to be numer-
stitial velocity of the solvent, is the void fraction of the bed, Dsf
ically integrated. These models have three and four adjustable
is the axial dispersion coefcient of the solute in the SCF, asf is the
parameters, respectively, which have to be tted to each experi-
effective soliduid contact area for mass transfer, Ksf is the mass
mental data set measured at constant temperature and pressure.
transfer coefcient for transport of the solute through the external
In this work we propose two simple models that have three and
uid lm around the solid particles, and Cf is the concentration of
two adjustable parameters, respectively, and explicitly express the
solute in the SCF lm that is in equilibrium with the solid surface.
relation between extraction yield and time; i.e., no differential
The balance equation is well known, and we use it here assuming
equations are present in the nal expression.
that the term that represents the axial dispersion (i.e., the rst term
on right hand side of Eq. (1) is negligible. According to Carberry
and Wendel [7], this approximation is valid when the length of the
extractor is at least 50 times the average particle diameter and the
Corresponding author. Tel.: +57 23 31 2935; fax: +57 23 39 2335. Reynolds number is greater than 10, which are conditions usually
E-mail addresses: camilo.pardo@correounivalle.edu.co (C. Pardo-Castano), accomplished in the SCF extraction of natural matter [8]. We also
gustavo.bolanos@correounivalle.edu.co (G. Bolanos). consider that the term that represents the solute accumulation in
http://dx.doi.org/10.1016/j.supu.2014.09.044
0896-8446/ 2014 Elsevier B.V. All rights reserved.
166 C. Pardo-Castano et al. / J. of Supercritical Fluids 97 (2015) 165173
(mo 0, xs ), we obtain an expression for as a function Eq. (12) expresses a relationship between time and extraction
of K: yield, and has three adjustable parameters: y*, K, and xm , which
2K K have a clear physical meaning. y* is related to the solubility of
= (10) the solute in the SCF corrected by diffusional limitations, K is the
2 (1 K)
ratio between the adsorption equilibrium constant of the solute
Because K 1 (and 0 1), only the root that is obtained with in the rst monolayer and that in subsequent layers (when the
the negative sign in the expression above is physically possible, and solutesolid interactions are strong, K , and when they are
thus in this case = 1. weak, K 0), and xm is the ratio between the mass of solute present
Now, in any case, replacing Eq. (9) in Eq. (6) we get: in the rst monolayer and the initial mass of solute that can be
extracted.
K 1+(xm /xs ) 2 + K 1 + xm /xs 2 2 4 (1 K) Notice that according to Eq. (22), for situations where diffusional
dxs
=
mf
y limitations are negligible (i.e., u, L , or k ), y* ysat , and
dt mo 2 (K 1) in these cases knowing the solubility of the solute in the SCF, the
number of adjustable parameters would reduce to two. On the other
(11) hand, note that xm , which is a parameter that represents properties
of the solid, might take values greater than 1 for cases where the
where y* is dened by Eq. (22) (see below). initial amount of solute is lower than the amount that can conform
the rst monolayer in the solid.
The argument of the exponential term in Eq. (22) is dimension-
2.4. Proposed models
less and can thus be expressed in terms of dimensionless numbers,
to give:
Eq. (11) can be analytically integrated using the initial condition:
at t = 0, xso = 1, to obtain, after several arrangements: Bi
y = ysat 1 exp (23)
mo ReSc
t= xo x + (2 K) x xm ln + Kxm ln (12)
2mf y (1 x) 2
where Bi, Re and Sc are the Biot, Reynolds and Schmidt numbers,
where: respectively. These numbers are dened as:
xo = a+b+c (13) hL KL
Bi = = (24)
2 D
x = a(1 x) + b (1 x) + c (14)
udp
a = K2 (15) Re = (25)
b = 2 (2 K) Kxm (16)
2
Sc = (26)
c = (Kxm ) (17) D
Fig. 2. Correlation of data for supercritical CO2 extraction of cocoa butter from cocoa powder. The numbers indicate the operating conditions from Table 1. () Proposed
model 1 (AARD = 3.81%), () Sovovs model (AARD = 3.02%).
This equation has only two adjustable parameters (y* and xm ), of the algorithm, no limit on the number of iterations was
and might be useful as a rst approximation for correlating exper- imposed.
imental data Figs. 2 and 3 show the correlation of data for the supercritical
CO2 extraction of cocoa butter from cocoa powder. Notice that both
3. Performance of the models and discussion proposed models apparently perform similarly, and they represent
the data adequately. Proposed model 1 looks like a more convenient
Some characteristics of the described models were initially choice because it represents adequately the whole set of data by
explored by applying them to the correlation of data obtained in using only one equation, in contrast to the Sovovs model, which
our laboratory for the supercritical CO2 extraction of cocoa butter uses three equations, one per period of extraction. Notice that the
from cocoa powder, a byproduct of the mechanical expression of AARD of the proposed model 1 (3.81%) compares well with that of
cocoa seeds. Extraction yields as a function of time were experi- Sovovs (3.02%).
mentally determined at ve sets of temperature and CO2 density Fig. 3 shows that the proposed model 2 and Sovovas model per-
[11,12]. The model proposed by Sovov [1], and also those pro- form similarly in the rst period of the extraction (i.e., in the linear
posed in this work were tted to these data, using a generalized part of the extraction yield vs. time curve). This is because in this
reduced gradient algorithm for nonlinear optimization, which was region external mass transfer domains, and both models include the
implemented in the Visual Basic solver for Excel 2013. The objec- solubility of the solute in the SCF, ysat . In contrast, at higher extrac-
tive function was dened as the average absolute relative deviation tion yields, important deviations occur for the proposed model 2
(AARD) referred to time, because this is the variable that is explicit probably because an internal mass transfer mechanism dominate
in the proposed models, and is expressed by the equation: (e.g. diffusion or desorption of the solute from the solid matrix) and
important simplifying assumptions related to this part were done
100 ti,exp ti,cal
n
during the derivation of proposed model 2. In fact, the deviation
AARD = ti,exp (28) obtained for the proposed model 2 was 5.75%, which is in contrast
n
i=1 with 3.02% that was obtained for Sovovs model. However, the
where AARD is the average absolute relative deviation (%), n is the proposed model 2 might be useful for providing rst estimates of
number of data (54), ti is time for data i, and subscripts exp and cal the y* and xm parameters for correlating experimental data with
refer to experimental and calculated, respectively. the proposed model 1.
In our computations we normalized time dividing this variable Table 1 shows the correlation of data on supercritical CO2
by the maximum experimental time. This avoided possible com- extraction of cocoa butter from cocoa powder. The values of the
putational problems that might arise from using absolute values of adjustable parameters were obtained by tting the models to the
time, and reduced the number of iterations required by the tting 54 experimental data points. For the proposed models, a single
routine. The initial values of the parameters for the proposed value of xm was obtained to represent the whole set of data because
model 1 were obtained by tting rst the proposed model 2 to the it was observed that this parameter does not strongly depend
data. The initial values for y* were numbers close to the solubility upon the operating conditions. In fact, this parameter is in essence
of the solute in the SCF (e.g., for cocoa butter is 0.45 wt%, which is a property of the solidsolute pair, so we did not expect a major
an approximated value according to the data presented by Saldana effect of the operating conditions on the value of this parameter. In
et al. [13]), and for the xm parameter we used 0.5. The routine was this case the values of xm were smaller than one, which indicates
allowed to run until the difference in the objective function calcu- that the initial mass of solute that is removable is greater than
lated in the last two iterations was 108 . Due to the fast convergence the mass of solute in the rst monolayer. In other cases (i.e., for
C. Pardo-Castano et al. / J. of Supercritical Fluids 97 (2015) 165173 169
Fig. 3. Correlation of data for supercritical CO2 extraction of cocoa butter from cocoa powder. The numbers indicate the operating conditions from Table 1. () Proposed
model 2 (AARD = 5.75%), () Sovovs model (AARD = 3.02%).
other systems) the opposite situation might be conceivable, and Table 2 and Fig. 4 show the correlation of the experimental data
the values of xm might be larger than one. for supercritical CO2 extraction of theobromine from cocoa seed
Notice that as the y* parameter increases, K decreases. Because shells at laboratory scale that were presented by Brunner [2]. Note
y* is directly related to solubility, when solubility of the solute in the that both proposed models perform similarly to the Sovovs and
SCF increases it is expected that the solidsolute molecular forces VT-II models for three of the four data sets reported at different
become relatively weaker, decreasing K. Reynolds numbers. During the correlation with the models, the K
For the Sovovs model, the three adjustable parameters (xk , Z, parameter was kept constant for the whole data set, as in this case
and W) were tted assuming that the values of Z and W depend temperature and pressure are constant and K is a parameter that
upon the operating conditions and that xk is independent of such represents the adsorption equilibrium.
variables. In fact, Z and W are directly related to the mass transfer Fig. 5 shows the same information than Fig. 4, except that in
coefcient, both in the SCF and in the solid matrix, respectively. this case the parameter K for the proposed model 1 was allowed
These coefcients must change with the extraction conditions. On to vary from one data set to another. Clearly, this improves the
the other hand, the xk parameter refers to the amount of solute in performance of the model to the point that despite the smaller
the solid matrix that is inaccessible to the SCF (i.e., xk = inaccessible number of parameters and the simplicity of its equation, the pro-
solute/initial solute), and thus it might be taken as a constant. posed model 1 ts the experimental data to an accuracy similar to
As a second example of the performance of the proposed mod- those of the other models. In fact, in this case, deviations (AARD) of
els, we present a comparison of results obtained with the Sovovs the proposed model 1, Sovovas and VT-II models were 1.97, 1.82,
model and the so-called VT-II model for correlating data on super- and 6.02%, respectively. These results indicate that the proposed
critical CO2 extraction of theobromine from cocoa seed shells model 1 compares favorably to the VT-II and Sovovs models.
that were presented by Brunner [2]. The VT-II model is also a Fig. 6 shows predictions by the proposed and VT-II models and
rigorous, phenomenological model that uses the Freundlich data for the supercritical CO2 extraction of theobromine from cocoa
isotherm to model the change in solute concentration in the solid. seed shells at pilot-plant scale. The deviations were 11.03% for VT-
If no axial dispersion is taken into account, it has four adjustable II, 12.52% for the proposed model 1, and 13.43% for the proposed
parameters. These are: two Freundlich coefcients, the effective model 2. Brunner [2] has shown that for this system the VT-II model
diffusivity of the extract in the solid, and a mass transfer coefcient can be used with reasonable accuracy for scaling purposes using the
in the uid. parameters obtained at laboratory scale. Although rigorous models
Table 1
Correlation of data on supercritical CO2 extraction of cocoa butter from cocoa powder.
CO2 density (g/cm3 ) Temperature ( C) Proposed model 1 Proposed model 2 Sovovs model
xm K y 10
* 5
xm y 10
* 5
xk Z 104 W 105
Table 2
Correlation of experimental data for supercritical CO2 extraction of theobromine from cocoa seed shells at laboratory scale. Experimental data by Brunner [2].
Fig. 4. Correlation of experimental data for supercritical CO2 extraction of theobromine from cocoa seed shells at laboratory scale. Proposed model 1 (AARD = 5.53%), proposed
model 2 (AARD = 6.83%), Sovovs (AARD = 1.82%) and VTII models (AARD = 6.03%). Experimental data at 30 MPa and 353 K [2].
Fig. 5. Correlation of data for supercritical CO2 extraction of theobromine from cocoa seed shells at laboratory scale. Proposed model 1 (AARD = 1.97%), Sovovs (AARD = 1.82%)
and VTII models (AARD = 6.03%). Experimental data at 30 MPa and 353 K [2]. Proposed model 1 when K is not constant.
C. Pardo-Castano et al. / J. of Supercritical Fluids 97 (2015) 165173 171
Fig. 6. Predictions and data obtained at pilot-plant scale for the supercritical CO2 extraction of theobromine from cocoa seed shells [2] by the proposed and VT-II models.
Fig. 7. Correlation of data for supercritical CO2 extraction of corn germ oil [14] by the proposed model 1. AARD = 3.89%.
Table 3
Parameters obtained for the proposed model 1 in correlation of data for supercritical CO2 extraction of three natural materials.
Temperature ( C) Pressure (MPa) CO2 velocity 105 (m/s) CO2 ow rate 104 (kg/s) xm K y* 103
Fig. 8. Correlation of data for supercritical CO2 extraction of Striped Weaksh oil [15] by the proposed model 1. AARD = 3.93%.
such as VT-II take into account parameters that can be affected at 20 MPa, 30 C and two solvent mass ow rates [15]. Fig. 9 shows
during scaling, we believe that as long as ow conditions such as the correlation of data for supercritical CO2 extraction of olive husk
Reynolds number are maintained the proposed models should be oil at 39.9 C and different pressures and solvent velocities [16].
as good (or bad) as more rigorous models for scaling purposes. Table 3 shows the parameters obtained for the proposed model 1
For the purpose of showing the performance of the proposed in correlation of data for these three systems.
model 1, which showed the highest accuracy in correlating the data Figs. 79 show that the performance of the proposed model 1 is
discussed above, we selected additional experimental data from the remarkable. This model is capable of correlating experimental data
literature and tted them with this model. Some of them are dis- for pressurized and supercritical extraction of natural matter to a
cussed here. Fig. 7 shows the correlation of experimental data for high accuracy, including not only vegetable matrices but also an
supercritical CO2 extraction of corn germ oil at 80 C and different animal matrix such as milled weaksh.
extraction pressures [14]. Fig. 8 shows the correlation of experi- Information presented in this paper illustrate the capability
mental data for pressurized CO2 extraction of striped weaksh oil that proposed model 1 has to correlate experimental data for
Fig. 9. Correlation of data for supercritical CO2 extraction of olive husk oil [16] at constant temperature of 40 C, by the proposed model 1. AARD = 7.28%.
C. Pardo-Castano et al. / J. of Supercritical Fluids 97 (2015) 165173 173
pressurized and supercritical extraction processes that experience is capable of representing the extraction periods as Sovovs model
changes in temperature (Fig. 2), pressure (Figs. 2, 7 and 9), and does, in some cases with higher accuracy. In addition, the proposed
Reynolds number (Figs. 4, 8 and 9). model 2 has a remarkably good performance considering that only
The proposed models are a close representation of the has one simple equation and two adjustable parameters. This model
phenomena that occur in a SCF extraction process within the frame- might be used as a rst approximation and can be used for gener-
work of the assumptions made, which are: (1) negligible axial ating rst estimates of the parameters for data correlation with the
dispersion, a condition that is obtained when the Reynolds number proposed model 1. The presented results indicate that the models
is greater than 10 and the length/diameter ratio of the extractor is proposed in this paper are simple and useful tools for correlat-
larger than 50; (2) the accumulation of solute in the SCF during the ing experimental data on supercritical uid extraction of natural
extraction is negligible when compared to the amount of the solute matter.
in the solid, which is attained at low values of the solubility of the
solute in the SCF, a usual situation for extraction of natural prod- Acknowledgments
ucts; (3) the concentration of the solute in the solid does not change
signicantly in the axial or radial directions, which is typically what The authors express their gratitude to Departamento Admin-
occurs in an industrial extractor, but not in a larger separator such istrativo de ciencia, tecnologa e innovacin de Colombia
as a chromatographic extractor; (4) a valid representation of the (COLCIENCIAS) for nancial support of this research (Grant 1106-
solutematrix interaction as an adsorption situation, which in an 454-21918).
extreme case might be attained when the extraction rate is not
limited by solubility, e.g., when the solute is located mainly intra- References
cellularly or the porosity of the solid matrix is small enough to limit
the extraction; and (5) only one solute is extracted by the SCF, or, [1] H. Sovov, Rate of the vegetable oil extraction with supercritical CO2 I. Mod-
elling of extraction curves, Chemical Engineering Science 49 (1994) 409414.
if several solutes are extracted, the difference in their solubilities is [2] G. Brunner, Gas Extraction, Springer-Verlag, Berlin Heidelberg, 1994, pp.
negligible or no interaction between the solutes is present (i.e., the 179250.
solubility of one solute in the SCF is not affected by the concentra- [3] H. Sovov, Mathematical model for supercritical uid extraction of natural
products and extraction curve evaluation, J. Supercritical Fluids 33 (2005)
tion of another one in the same phase). We believe that many cases
3552.
in SCF extraction of natural matter correspond to these approxima- [4] H. Sovov, Steps of supercritical uid extraction of natural products and their
tions. Clearly, as the characteristics of the system move away from characteristic times, J. Supercritical Fluids 66 (2012) 7379.
these assumptions, the models should be considered as a useful [5] J.M. del Valle, J.C. de la Fuente, Supercritical CO2 extraction of oilseeds: review
of kinetic and equilibrium models, Critical Review in Food Science and Nutrition
semiempirical approximation to the process. 46 (2006) 131160.
Adsorption-based models for supercritical extraction of natu- [6] C. Grosso, J.P. Coelho, F.L.P. Pessoa, J.M.N.A. Fareleira, J.G. Barroso, J.S. Urieta,
ral matter have been in the literature for some time. In particular, A.F. Palavra, H. Sovov, Mathematical modelling of supercritical CO2 extraction
of volatile oils from aromatic plants, Chemical Engineering Science 65 (2010)
several models are known that use the BET approach to model 35793590.
adsorption equilibrium together with rigorous balances [10,17] to [7] J.J. Carberry, M.M. Wendel, A computer model of the xed bed catalytic reactor:
take into account diffusion in micropores, mass transfer to the the adiabatic and quasi-adiabatic cases, AIChE J. 9 (1963) 129133.
[8] Z. Huang, X.H. Shi, W.J. Jiang, Theoretical models for supercritical uid extrac-
supercritical uid, and radial and axial dispersion in the extrac- tion, J. Chromatography A 1250 (2012) 226.
tor. Simple models such as those presented in this work capture [9] S. Brunauer, P.H. Emmett, E. Teller, Adsorption of gases in multimolecular lay-
essential elements of the extraction. For example, in addition to the ers, J. American Chemical Society 60 (1938) 309319.
[10] M. Goto, B.C. Roy, A. Kodama, T. Hirose, Modeling supercritical uid extraction
adequate correlation of data that was shown, the models correctly process involving solutesolid interaction, J. Chemical Engineering of Japan 31
predict that the apparent solubility (i.e., the slope of the extraction (1998) 171177.
yield vs. time curve) is a function of the amount of solute initially [11] I. Meja, G. Bolanos, Supercritical uid extraction of cocoa butter: a promising
technology for adding value to an economically important natural product, in:
present in the bed, an observation that has been presented in the
S. Bottini, S. Pereda (eds.), Book of abstracts, I Iberoamerican Conference on
literature using more elaborated analyses [18]. The models can thus Supercritical Fluids, Prosciba, Iguazu Falls, Argentina and Brazil, April 1013,
be useful in studies on supercritical extraction of natural matter. 2007, p. 128.
[12] M. Velasquez, Studies on the Extraction of Cocoa Butter from Cocoa Powder
Using Supercritical Carbon Dioxide, Universidad del Valle, Cali, Colombia, 2011
4. Conclusions (M.Sc. Thesis in Chemical Engineering).
[13] M.D.A. Saldana, R.S. Mohamed, P. Mazzafera, Extraction of cocoa butter from
Two simple mathematical models that adequately correlate Brazilian cocoa beans using supercritical CO2 and ethane, Fluid Phase Equilibria
194197 (2002) 885894.
information on the kinetics of supercritical uid extraction were [14] S. Rebolleda, N. Rubio, S. Beltrn, M.T. Sanz, M.L. Gonzlez-Sanjos, Supercritical
presented. The models integrate modications of the BET adsorp- uid extraction of corn germ oil: study of the inuence of process parame-
tion equilibrium equation with a simplied material balance of ters on the extraction yield and oil quality, J. Supercritical Fluids 72 (2012)
270277.
a supercritical extractor. The new models have three and two [15] A.C. Aguiar, J.V. Visentainer, J. Martnez, Extraction from striped weaksh
adjustable parameters, respectively, each one with a clear physi- (Cynoscion striatus) wastes with pressurized CO2 : global yield, composition,
cal meaning. These models produce an average deviation similar kinetics and cost estimation, J. Supercritical Fluids 71 (2012) 110.
[16] M.M. Esquvel, M.G. Bernardo-Gil, M.B. King, Mathematical models for super-
to that obtained with rigorous models such as Sovovs and the critical extraction of olive husk oil, J. Supercritical Fluids 16 (1999) 4358.
so-called VT-II models. In particular, the performance of proposed [17] L. Ruetsch, J. Daghero, M. Mattea, Supercritical extraction of solid matrices.
model 1 is in several cases better than those of the VT-II model, a Model formulation and experiments, Latin American Applied Research 33
(2003) 103107.
remarkable result given that it is only constituted by one equation, [18] J.M. del Valle, F.A. Urrego, Free solute content and solutematrix interactions
as compared to three conditional equations of Sovovs model and affect apparent solubility and apparent solute content in supercritical CO2
the system of differential equations of VT-II. The proposed model 1 extractions. A hypothesis paper, J. Supercritical Fluids 66 (2012) 157175.