Annatto seeds of the variety Piave (Bixa orellana L.) were donated by Urucum do Brasil Ltda, located in Santa Marta, Monte Castelo - SP, Brazil. The seeds were stored at 291 K in a freezer (VF55D, Metalfrio, São Paulo, Brazil). The mean diameter of the annatto seeds, 3.82 ± 0.09 mm, was calculated by an adaptation of Silva et al. [19Silva GF, Gamarra FMC, Oliveira AL, Cabral FA. Extraction of bixin from annatto seeds using supercritical carbon dioxide. Braz J Chem Eng 2008; 25: 419-26.
[http://dx.doi.org/10.1590/S0104-66322008000200019] ]. A total of 30 annatto seeds were removed from 3.5 kg of seeds for the measurement of the lengths of their axes. Then, to repeat the measurement of lengths, 30 seeds were placed back into the bag (which again weighed 3.5 kg), and after homogenization, the procedure was repeated.

The moisture content of the annatto seeds, 11.61 ± 0.02 g/100 g, was analyzed as described by AOAC [20AOAC. Official Methods of Analysis of AOAC International 1997.] and was determined in triplicate by drying at 378 K until reaching constant mass. The true density (1.35 ± 0.01 g/cm³) of the particles was measured by pycnometry with helium gas at the Central Analytical Laboratory of the Institute of Chemistry - UNICAMP (Campinas, Brazil). The apparent density (0.7609 ± 0.0000 g/cm³) in SFE bench-scale unit, of the bed was calculated by dividing the sample feed mass by the extractor volume. The porosity of the bed was calculated by equation 1.

(1)

where ρ_r is the true density of the particles; ρ_a is the apparent density of the bed; and ε: is the porosity of the bed.

Two commercial supercritical fluid extraction units were used in this study, a bench-scale unit and a pilot-scale unit. In the bench-scale unit, the extraction kinetics of annatto seeds was investigated. However, in the pilot-scale unit, several elements were changed and adapted to test the possibility of recycling CO₂ without creating drag or with the least amount of drag, which would prevent the CO₂ from entering the extraction column.

2.2.1. SFE Bench-scale Unit

The main elements that compose the commercial SFE bench-scale unit (Spe-ed model, Applied Separations, Allentown, USA) are represented in Fig. (1). The extraction column used in this study has a total volume of 25 mL, height of 83.5 mm, and diameter of 20.0 mm.

Fig. (1) SFE bench-scale unit. (1) CO2 cylinder, (2) blocking valve, (2’) safety valve, (3) line filter, (3’) air filter, (4) cooling bath, (5) compressor, (6) air pressure regulator, (7) air-driven fluid pump, (8) manometer, (8’) temperature indicator, (9) blow motor, (10) blow motor resistor, (11) extraction column, (12) oven module, (13) micrometric valve with heating system, (14) collection flask, (15) flow meter, (16) flow totalizer.

The operational sequence was as follows: CO₂ was cooled at 263 K by a cooling bath (Logen Scientific, LS-540, Diadema, Brazil) before reaching the pump. The desired temperature for CO₂ was obtained using hot air circulation by blow motor and blow motor resistor inside the oven module, which included the extraction column and the pipes connected to the inlet and outlet of the column. Accordingly, CO₂ transited a thermodynamic phase change from liquid state to supercritical state. Then, the extractor was kept at 20 MPa and 313 K for 20 min (static time) to get saturation of the solvent in the extractor before solvent pumping. After that, the outlet valve of extractor was opened, and the dynamic time of extraction was started. The extract was collected in a 100 mL glass vial immersed in an ice bath at ambient pressure. The total CO₂ was measured with a flow totalizer (Itrón, model G1.0, Campinas, Brazil), and the mass flow for each process was calculated in an Excel spreadsheet, weighting the values of the density of the CO₂ found elsewhere [21 NIST. National Institute of Standards and Technology. 2017. Available at: http://www.nist.gov] keeping the environment at a constant temperature; CO₂ was not recirculated. After the extraction, the glass vial was placed in a desiccator for thirty minutes to guarantee the complete elimination of CO₂.

2.2.2. SFE Pilot-scale Unit Description

The commercial SFE pilot-scale unit (Thar Technologies, model SFE-2X5 LF-2-FMC, Pittsburgh, USA) used in this study was acquired in the framework of the PITE/FAPESP project (05/60948-1) in partnership with the company CENTROFLORA (Botucatu, SP) during the period 2006-2009.

The main elements that compose the initial state of the commercial SFE pilot-scale unit are represented in Fig. (2). The extraction column used in this study has a total volume of 5 L, height of 620 mm, and diameter of 106 mm. The open-closure sealing system is an internal thread with a depth of 7 threads x 39.0 mm and the cover is the external thread with a depth of 6 threads x 36.0 mm and sealing gasket (the maximum operating pressure and temperature are 68.95 MPa and 363 K, respectively).

Fig. (2) Initial state of SFE pilot-scale unit. (1) CO2 cylinder, (2) blocking valve, (2’) safety valve, (3) line filter, (4) manometer, (4’) internal temperature indicator, (4”) external temperature indicator, (5) heat exchanger with coiled tube for CO2 cooling, (6) mass flow meter, (7) CO2 pump, (7’) cosolvent pump, (8) heat exchanger for CO2 heating, (9) three-way valve, (10) extraction column, (11) automated backpressure valve, (11’) manual backpressure valve, (12) separation vessel.

The operational sequence was as follows: CO₂ passed through the heat exchanger with coiled tube where it was cooled at 263 K by a cooling bath (Polyscience recirculator 6100T, Niles, Illinois, USA) and for the mass flow meter (Siemens Coriolis flowmeters SITRANS F C MASS 2100 Di 1.5), before reaching the pump (maximum flowrate of 350 mL/min, maximum operating pressure of 34.47 MPa). Then CO₂ passed through a heat exchanger, which raised the temperature and entered into the extraction column equipped with an electric heating jacket (Ogden Mighty-Miser CBE06A12A-02212, 2000 W, 240 V) kept to 313 K and 20 MPa for 20 min (static time) to get saturation of the solvent in the extractor before solvent pumping. After that, outlet three-way valve of extractor and the automated backpressure valve were opened, and the dynamic time of extraction was started. The extract was collected in three separators (Cyclone type, volume of 1 L, diameter of 74 mm, height of 275 mm. Each one with an electric heating jacket, Ogden Mighty-Miser CBE04J04A-02158, 500 W, 240 V). The total CO₂ was measured with a flow totalizer (Itrón, model G1.6, Campinas, Brazil), and the mass flow for each process was calculated in an Excel spreadsheet, weighting the values of the density of the CO₂ found elsewhere [21 NIST. National Institute of Standards and Technology. 2017. Available at: http://www.nist.gov] keeping the environment at a constant temperature. CO₂ was recirculated in 4 of 5 experiments (B, C, D and E). After the extraction, the extract was collected in glass bottles of 100 mL and placed in a desiccator for thirty minutes to guarantee the complete elimination of CO₂.

To recirculate CO₂ at purity conditions similar to those of the initial feed, an AdC was installed between the SV3 outlet and the CO₂ condenser, which was connected to what is here called the lung tank (LT). The capacity of LT is 15 L volume, having a maximum allowable working pressure of 10.34 MPa at 298 K, and it is made of 304 stainless steel.

Fig. (3) shows the initial step of the recycle system construction, as well as its main components.

Fig. (3) Initial step of the recycle system construction. (2) blocking valve, (2’) safety valve, (4) manometer, (16) shell and tube heat exchanger, (17) lung tank (LT).

The purpose of installing an AdC (this column was maintained at an average room temperature of 295 K, it has no heating system) as part of the initial CO₂ recycle system was to keep the CO₂ mass in the same condition as if it were entering the system for the first time (i.e., 99.9% purity). To this end, a 0.65 L (height of 22 cm and diameter of 12 cm) AdC was installed with a set of connections and valves that complemented the pilot-scale unit’s recycle system. Fig. (4) shows the sequence of connections, valves, and piping that was installed in the final stage of the study, and Fig. (5) shows the final state of the SFE pilot-scale unit after the inclusion of the online subcritical adsorption-based device (shown in red).

Fig. (4) Final step of the recycle system construction. Adsorption column (AdC) and other components as part of the recycle system of the pilot-scale unit. (2) blocking valve, (6) diaphragm flow meter, (AdC) adsorption column, (14) heat exchanger with electric resistance to heat the micrometering valve, (15) micrometering valve.

Fig. (5) Final state of SFE pilot-scale unit after online subcritical adsorption-based device inclusion. (1) CO2 cylinder, (2) blocking valve, (2’) safety valve, (3) line filter, (4) manometer, (4’) internal temperature indicator, (4”) external temperature indicator, (5) heat exchanger with coiled tube for CO2 cooling, (6) mass flow meter, (7) CO2 pump, (7’) cosolvent pump, (8) heat exchanger for CO2 heating, (9) three-way valve, (10) extraction column, (11) automated backpressure valve, (11’) manual backpressure valve, (12) separation vessel, (13) adsorption column (AdC), (14) heat exchanger with electric resistance, (15) micrometering valve, (16) shell and tube condenser, (17) lung tank (LT).

To facilitate cleaning and to understand in detail the parts that composed the developed recycle device, all the connections were disassembled to evaluate the possibility of avoiding dragging the extract toward the LT (17, in Fig. (3)). The blocking valve located at the inlet of the refrigeration heat exchanger (CO₂ condensing column) can operate at up to 41.37 MPa CWP (cold working pressure) and 311 K. The shell and tube heat exchanger used for refrigeration was of model # 00256-06 (Garden City, NY, USA), and the tube side pressure could withstand maximum conditions of 8.3 MPa [28 EXERGY-LLC. Model # 00256-06, Shell & Tube Heat Exchanger. USA, 17/05/2017 2017. Available at: http://heat-exchangers. exergyllc.com/item/shell-tube-heat-exchangers/35-series-shell-tube-heat- exchangers/ 00256-06]. The shell and tube heat exchanger contains 61 numbers of tubes with length and diameter of 51 cm, and 2.5 mm respectively. The cooling solution, composed of water + ethylene glycol (50:50 v/v), entered the side of the heat exchanger by applying pressure to the inner pipes of the exchanger. According to PolyScience [29 POLYSCIENCE. Chillers and Coolers. USA, 2017. Available at: https://polyscience.com/products/chillers/benchtop], the maximum pressure exerted by the recirculating cooling bath (PolyScience recirculator 6100T - 230 V / 60 Hz / 12.2 A / 1 HP, Niles, Illinois, USA) that drives the refrigerant solution is 0.62 MPa. In addition, the minimum range of the flow rate alarm set point is 3 to 8 LPM (liters per minute), and the maximum flow rate at 0 MPa is 13.2 LPM [30POLYSCIENCE. User Guide - Refrigerated Recirculating Chillers 2016; 51.].

With the modifications made in the recycling system, the experiments from B to E were developed as follows: the processes B and C are as follows: 1) The extractor was pressurized from 0.1 MPa to 20 MPa during 20 min. Then, the extractor remains in this condition for 20 min (static time). After that, the extractor was depressurized to 7 MPa. 2) The extractor was pressurized again from 7 MPa to 20 MPa during 10 min and remained in this condition for 20 min (static time). Then, the output valve of extractor was opened to start the dynamic extraction until reaching values of S/F ratio (g CO₂/g annatto) 25, for B, and 10.36, for C. Next, the recycling valve (2d in Fig. (5)) was closed, and the ambient valve (2l in Fig. (5)) was opened to flow out the CO₂ to the ambient. After that, the extract was collected and weighted. 3) The same procedure of part 2 was repeated except that the pressurization occurs from 0.1 MPa to 20 MPa during 20 min.

In cases D and E, first the extractor was pressurized from 0.1 MPa to 20 MPa during 20 min. Then, the extractor remains in this condition for 20 min (static time). After that, the output valve of extractor was opened to start the dynamic extraction until reaching values of S/F 25.45, for D, and 55.98, for E. Next, the recycling valve (2d in Fig. (5)) was closed, and the ambient valve (2l in Fig. (5)) was opened to flow out the CO₂ to the ambient.

The global extraction yield (X0) was obtained from kinetic curves corresponding to the diffusion-controlled period. In practical aspects, the global extraction yield (X0) represents the maximum amount of extract that can be recovered from a raw material at a given extraction pressure and temperature. In contrast to the bench-scale unit, in the pilot-scale unit, it is not possible to collect extracts continuously during kinetics experiments. It is necessary to interrupt the solvent flow and then depressurize the separators while keeping the extractor pressurized so that the extract can be collected, as is the case with experimental runs B and C. After this procedure, the extraction can be resumed until the next collection point. Interrupting the extraction at each point implies intermediary static periods in the extractor. Thus, in this study, we used the bench-scale unit to determine the overall extraction curve (OEC) and X0 for annatto-seed extraction.

Fig. (6) shows the OEC obtained for annatto-seed SFE at the bench scale. The results are expressed in terms of the relative yield (R, g extract/100 g of extractable) of the extraction. The term extractable represents the maximum amount of extract that can be recovered from a raw material at a given extraction pressure and temperature within the diffusion-controlled period (S/F ratio of 259 g CO₂/g annatto, which is in a good agreement with the value obtained by Moraes et al. [26Moraes MN, Zabot GL, Meireles MAA. Extraction of tocotrienols from annatto seeds by a pseudo continuously operated SFE process integrated with low-pressure solvent extraction for bixin production. J Supercrit Fluids 2015; 96: 262-71.
[http://dx.doi.org/10.1016/j.supflu.2014.09.007] ]). The OEC presents a short CER (constant extraction rate) period of 26.17 min, which accounts for almost 60% of the total extract. A DC (diffusion controlled) period follows, with a rapid decrease in the extraction rate; this is the intermediary FER (falling extraction rate) period. The kinetic parameters calculated from the analysis performed using the software SAS 9.2^® are in Table 1, where t_CER is the time of the CER, M_CER is the mass-transfer rate and R_CER is the yield achieved during this period. The value of t_CER roughly represents the minimum time an SFE cycle should last to keep the process economically viable [25Meireles MAA. Extraction of bioactive compounds from latim american plants.Supercritical Fluid Extraction of Nutraceuticals and Bioactive Compounds: CRC Press - Taylor & Francis Group, LLC 2008; 243-74.]. Y_CER is the mass ratio of solute in the supercritical phase at the bed outlet during the CER period.

Fig. (6) Overall Extraction Curve (OEC) for annatto-seed extraction (313 K and 20 MPa) in the bench-scale unit; (__) indicates data fitted using SAS 9.2®.

Table 1
Kinetic parameters of SFE from annatto seeds estimated using an SFE bench-scale unit*.

Obtaining the CO₂ density as a function of ambient pressure (from [21 NIST. National Institute of Standards and Technology. 2017. Available at: http://www.nist.gov]) together with the monitoring of both ambient pressure (measured through an aneroid barometer) and the volumetric flow of CO₂ (measured with the flow totalizer -16 of Fig. (1) and with a chronometer), allowed us to control CO₂ mass flow rate as accurate as possible. Accordingly, the error for CO₂ flow rate, 0.04 g/min, compared to the average of CO₂ flow rate, 11.10 g/min, indicates the infinitesimal impact of the precision of CO₂ flow rate on both the overall extraction curves and model parameters in this study.

Based on the OEC and kinetic parameters calculated, we also performed 3 runs in the SFE bench-scale system to ensure the reliability of experimental data using different solvent mass to feed mass (S/F) ratios which can be seen in Fig. (6). The Table 2 shows the results for the global extraction yield (X0) and residual extract indicators when S/F was varied. It can be observed that both the S/F ratio and the global extraction yield (X0) values had a coefficient of variation close to 1%, which represents a high reliability in the performed process. The Standard deviation reported in this table are related to the uncertainty of the experiments.

Table 2
Percentage representation of the total extract and the residual extract for defatting annatto seeds using the bench-scale unit.

On the other hand, the process had a high variability with respect to the values obtained related to the extract collected during the cleaning of the unit (residual extract amount), which can be seen in terms of the standard deviation and coefficient of variation, indicating a strong correlation to the S/F ratio. The lower the S/F ratio is, the lower is the X0 and the higher is the accumulation of extract inside the parts of the SFE bench-scale unit, and vice versa. In contrast, approximately 6% of the total amount of extract that can be extracted from the annatto seeds using the SFE conditions used here (313 K and 20 MPa) corresponds to the amount of extract lost from the retention of extract in the pieces of equipment used. The residence time of CO₂ in the extractor of bench unit was 0.08 s, while the residence time of pilot unit was from 8.03 to 19.60.

Santos et al. [31Santos DT, Albuquerque CLC, Meireles MAA. Procedia Food Sci 2011; 1: 1581-8.
[http://dx.doi.org/10.1016/j.profoo.2011.09.234] ], using the same commercial equipment, observed a similar percentage of residual extract (7%) for annatto-seed extraction under different SFE conditions (31 MPa, 333 K). The authors also compared commercial equipment with a homemade system that had a similar volume capacity and observed that a significantly higher amount of residual extract was deposited in the tubing line when the homemade equipment was used compared with the amount of residual extract obtained using the commercial SFE apparatus; thus, when using the homemade modification that we applied to the pilot-scale unit, accounting for this loss of extract was a concern.

Before studying the effects of the ratio of solvent mass to feed mass on the extraction yield using the SFE pilot-scale unit, we performed preliminary tests with the developed online subcritical adsorption-based device coupled to the system to validate the device. Silicon dioxide (Celite^® 512 medium) and cotton (in a uniform orthopedic blanket roll) (Hydrophilic Cotton, Medi House, São Paulo, Brazil) were tested as adsorbent materials. The technical viability validation criterion adopted, in addition to visual observation of the adsorbent material after the SFE process, consisted of monitoring the effects of the increased solvent mass to feed mass (S/F) ratio on the extraction yield and percentage of residual extract in the equipment parts and comparing the results for the SFE pilot-scale unit coupled with the online subcritical adsorption-based device with the results for the SFE bench-scale unit.

Table 3 shows the results for extraction yield and residual extract percentage when solvent mass to feed mass ratio (S/F) was varied. The term extraction yield is defined as the summation of mass of extract obtained in three separators per 100 g of raw material. Moreover, the term residual extract percentage in this table indicates the amount of residual extract retained inside the tubing system of the SFE unit. Based on this table, the Extraction yield was increased from 2.84, for B, to 3.09, for C. The residual extract was decreased from 8.12, for B, to 6.55, for C. The main reason for this drop of residual extract is the higher value of S/F for C compared to B that leads higher removal of extract available in the tube lines and connections. It can be observed that the values obtained related to the extract collected during the cleaning of the unit (residual extract percentage) presented a strong correlation to the values of S/F ratio.

Table 3
Percentage representation of the total extract and the residual extract for defatting annatto seeds using a pilot-scale unit with a coupled subcritical adsorption-based device.

The higher the S/F ratio, the higher the extraction yield (measured by collecting the extracts in the 3 separation vessels), and the lower the accumulation of extract inside the parts of the SFE pilot scale unit (Residual Extract Percentage %). So, the use of silicon dioxide as adsorbent material for increasing extract recovery besides allow CO₂ recycling is an important issue to be addressed.

Fig. (7) Adsorbent materials used in the online subcritical adsorption-based device. B and C are Celite, and D is cotton. D1, D2 and D3 correspond to successive layers of cotton after removing the first layer seen in D.

The two adsorbent materials tested were compared in terms of operational aspects and cost. Because cotton is a soft-textured fibrous material, it facilitated the operability of the adsorption column at the time of packaging and unpacking. Almost 100 g (with a cost equivalent to approximately R$ 25.00/kg, which is very cheap - approximately 10% of the cost of silicon dioxide (Celite^® 512 medium, R$ 263.00/kg)) was adequate to fill the 0.65 L volume of the adsorption column. In addition, other operational limitations (i.e., the formation of preferred paths, as well as the loss of mass between the internal spaces of the adsorption column covers when collecting the Celite^® 512 medium at the end of processing), made this product less preferable compared with cotton.

Another important conclusion regarding the operability of the proposed supercritical fluid extraction process concerned the cleaning step.

The need to reduce these cleaning times and the amount of cleaning solvents required led to the identification of empty spaces (Fig. 8) where the extracted material accumulated after each processing run. Studies performed by Torres [32Torres RADC. Purificação de partículas de bixina por precipitação com CO2 supercrítico como antissolvente a partir de solução de ativos obtida da extração de sementes semi desengorduradas de urucum (Bixa orellana L.) assistida por ultrassom. Master thesis, University of Campinas/School of Food Engineering, Campinas, Brazil, 2015.] showed a similarity to this case, and therefore, the same strategies, such as the use of Teflon bushings, were applied to improve the equipment. Using Teflon bushings in the separators caused a dramatic drop in both cleaning time, from 40 h (after each experimental run, especially in cases of A. B and C) to 12 h (after each experimental run, D and E), and consumption solvent (essentially ethanol) volumes, from 14 L to 3 L.

Fig. (8) Teflon bushings used to avoid the accumulation of extract in the SFE pilot-scale unit coupled with the developed online subcritical adsorption-based device.

After unit adjustments and the technical viability demonstration of the inclusion of the subcritical adsorption-based device in the SFE pilot-scale unit, 5 experimental runs (A-E) were performed to evaluate the effect of the solvent mass to feed mass (S/F) ratio on the extraction yield to validate the technical viability and better understand the process. Based on the OEC and kinetic parameters calculated using the data from defatting annatto seeds using the SFE bench-scale system, the extraction conditions were selected.

Run A was performed according to the initial state of the pilot-scale unit, i.e., without any changes being made to the equipment. Thus, this extraction process served to allow data such as CO₂ consumption (approximately 13 kg) to be surveyed under the processing conditions. In addition, Run A was performed to better compare the effects of the inclusion of the adsorption column and Lung Tank (LT) in assisting CO₂ recycling. Runs B and C correspond to the tests in which silicon dioxide (Celite^® 512 medium) was used as an adsorbent material. Runs D and E correspond to the tests where the cotton (in a uniform orthopedic blanket roll) was used as an adsorbent material.

Table 3 shows the results for extraction yield and residual extract indicators when S/F was varied. It should be observed here that the extraction yield differs from the global extraction yield (X0) previously obtained in the bench-scale unit since the latter represents the maximum amount of extract that can be recovered from a raw material at a given extraction pressure and temperature (2.82% for annatto seeds for SFE at 313 K and 20 MPa). As observed for the bench-scale unit experiments, the values obtained related to the extract collected during the cleaning of the unit (residual extract amount) were strongly correlated to the S/F ratio. The higher the S/F ratio is, the higher is the extraction yield measured by collecting the extracts in the 3 separation vessels and the lower is the accumulation of extract inside the parts of the SFE pilot-scale unit. Comparing the percentages of residual extract obtained in the pilot-scale unit during runs A-E (7.58-16.65%) with those obtained previously in the bench-scale unit (4.05-8.23%), it can be observed that the pilot-scale unit retains more extract in its tubing (7.58-16.65%). Therefore, any gain related to an increase in the amount of extract collected would be interesting. Therefore, the use of cotton as an adsorbent material for increasing extract recovery in addition to allowing CO₂ recycling is a critical issue to be addressed.

In this context, we can see that an increase of 3.8 and 1.6% in the extraction yield was obtained using cotton inside the online subcritical adsorption-based device in runs D and E, respectively. In terms of the adsorbed extract amount (in grams) divided by 100 g of cotton, equivalent values of 1.02% and 0.67% were obtained, indicating that the amount of annatto-seed oil in the final adsorbent material is approximately 1% (see Fig. 7d, the colored cotton material with extract). Considering that in Run E, the S/F ratio was approximately double the value applied in Run D, we can infer that similarly to what occurred with the residual extract deposited in the tubing lines in both units tested here, the use of a higher S/F ratio is also preferred for reducing extract accumulation inside the adsorption column. In our case, when the S/F ratio was increased by 119.61%, the amount of adsorbed extract was reduced by 34.31%. Table 3 also shows the improvement in terms of the lower retention and accumulation of extracts obtained by the inclusion of Teflon bushings in the unit, comparing runs A, B and C (without unit adjustments) with runs D and E (with Teflon bushings and coupled with the developed online subcritical adsorption-based device.

Del Valle et al. [33Del Valle JM, Rivera O, Mattea M, Ruetsch L, Daghero J, Flores A. Supercritical CO2 processing of pretreated rosehip seeds: effect of process scale on oil extraction kinetics. J Supercrit Fluids 2004; 31(2): 159-74.
[http://dx.doi.org/10.1016/j.supflu.2003.11.005] ] mentioned that one of the reasons for a low extraction yield compared with that obtained on an analytical (or laboratory) scale is the dispersion of the material extracted between the extraction column and the

separators. Corroborating these findings, in this study, we found that 92.54% and 94.37% of the total residual extract was provided by the separation-vessel cleaning step in runs B and C, respectively.

About the possibility of channeling in this study, it should be noted that the apparent density of the raw materials inside the extractor was the same for the whole runs, 0.7 g/cm3. According to the reference 33Del Valle JM, Rivera O, Mattea M, Ruetsch L, Daghero J, Flores A. Supercritical CO2 processing of pretreated rosehip seeds: effect of process scale on oil extraction kinetics. J Supercrit Fluids 2004; 31(2): 159-74.
[http://dx.doi.org/10.1016/j.supflu.2003.11.005] , the possibility of channeling is higher for a bed with height per diameter ratio of unity, which is far away from height per diameter ratio of 2.4 to 5.8 in the current study. Homogeneous color of the raw material at the end of the extraction is another reason that CO₂ distributed homogeneously inside the bed and channeling problem did not occur.

Table 3 shows that increasing S/F improves the extraction yield (g extract/ 100 g annatto seeds) due to higher CO₂ consumption. This table also reports the adsorbed mass of extract that is 1 g, for Run B with S/F of 25.45, and 0.66 g, for Run C with S/F of 55.98. One probable reason for lower amount of adsorbed mass of extract of Run C compared to run B is the higher amount of S/F in Run C that carries out higher amount of the extract available in the adsorption column. Accordingly, the total extraction yield, which includes the mass of extract inside the separators and the mass of extract in the adsorption column, determined to be 1.42, for Run A, 1.89, for Run B, and 2.46, for Run C. Moreover, as shown in the last column of this table, an increment of S/F from 3.69 to 55.98 reduces the amount of extract in the lines from 16.65% to 7.58%. This is mainly due to the employing adsorption-base device together with Teflon bushings in the separators that leads to recovering higher amount of extract and reducing the amount of extract in the lines.

Considering the need to perform tests in the pilot-scale unit to verify the possibility of recirculating CO₂ with the least amount of drag on the extract and without requiring a great initial investment, some components were borrowed from other units: a column that was used as adsorption column, a manometer and a micrometering valve (Fig. 9). Then, corrective maintenance of 4 valves that were previously in disuse was performed; these valves appear with a (*) in Table 4. In this sense, disregarding the cost calculation of items with (*), we can say that an initial investment of only USD 461.36 was necessary.

Fig. (9) Components of the homemade online subcritical adsorption-based device developed to allow effective CO2 recycling.

Table 4
Estimated cost of the components of the homemade device developed to allow effective CO₂ recycling.

On the other hand, an estimated total cost of USD 5731.36 would be required to build the developed device in an SFE unit like the one used here.